CN109729171B - Method for constructing town cognitive matrix Internet of things - Google Patents
Method for constructing town cognitive matrix Internet of things Download PDFInfo
- Publication number
- CN109729171B CN109729171B CN201910022071.4A CN201910022071A CN109729171B CN 109729171 B CN109729171 B CN 109729171B CN 201910022071 A CN201910022071 A CN 201910022071A CN 109729171 B CN109729171 B CN 109729171B
- Authority
- CN
- China
- Prior art keywords
- data
- learning
- convolution
- training
- time
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 239000011159 matrix material Substances 0.000 title claims abstract description 104
- 238000000034 method Methods 0.000 title claims abstract description 70
- 230000001149 cognitive effect Effects 0.000 title abstract description 27
- 238000012549 training Methods 0.000 claims abstract description 138
- 230000006399 behavior Effects 0.000 claims abstract description 130
- 239000000523 sample Substances 0.000 claims abstract description 128
- 238000002372 labelling Methods 0.000 claims abstract description 74
- 238000010276 construction Methods 0.000 claims abstract description 70
- 238000013528 artificial neural network Methods 0.000 claims description 90
- 238000012360 testing method Methods 0.000 claims description 89
- 238000004364 calculation method Methods 0.000 claims description 73
- 238000010801 machine learning Methods 0.000 claims description 70
- 230000000694 effects Effects 0.000 claims description 60
- 230000008569 process Effects 0.000 claims description 50
- 238000006243 chemical reaction Methods 0.000 claims description 48
- 238000001514 detection method Methods 0.000 claims description 47
- 238000005457 optimization Methods 0.000 claims description 29
- 238000004458 analytical method Methods 0.000 claims description 28
- 238000005516 engineering process Methods 0.000 claims description 27
- 238000005259 measurement Methods 0.000 claims description 24
- 238000012800 visualization Methods 0.000 claims description 19
- 238000012546 transfer Methods 0.000 claims description 14
- 230000008859 change Effects 0.000 claims description 12
- 238000007405 data analysis Methods 0.000 claims description 11
- 230000019771 cognition Effects 0.000 claims description 8
- 238000013461 design Methods 0.000 claims description 8
- 238000013021 overheating Methods 0.000 claims description 8
- 238000011176 pooling Methods 0.000 claims description 7
- 239000003086 colorant Substances 0.000 claims description 5
- 230000021824 exploration behavior Effects 0.000 claims description 5
- 239000000470 constituent Substances 0.000 claims description 3
- 241000287107 Passer Species 0.000 description 4
- 238000011161 development Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 230000002452 interceptive effect Effects 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 230000002596 correlated effect Effects 0.000 description 2
- 230000000875 corresponding effect Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 1
- 230000003920 cognitive function Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 230000004083 survival effect Effects 0.000 description 1
Images
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a method for constructing a town cognitive matrix Internet of things, which comprises the following steps: s1, carrying out location association construction on the composition elements in the region based on a knowledge graph; s2, constructing a probe grid for detecting the mobile equipment in the area; s3, performing matrix acquisition construction for acquiring the mobile data of the mobile equipment in the area, and performing data labeling construction for labeling the mobile data; and S4, carrying out topology convolution construction for machine training and learning and carrying out prediction classification construction based on the mobile data. The internet of things constructed by the invention has the cognitive capabilities of identifying and understanding behavior track characteristics of mobile phone holders such as small-town customers, passerby and visitor, and self-managing, self-learning and self-optimizing.
Description
Technical Field
The invention relates to a construction method of an Internet of things, in particular to a construction method of a small town cognitive matrix Internet of things.
Background
With the development of the times, the aspects of people in daily life become more intelligent and automatic. Therefore, the technology of the internet of things is getting more and more attention from the beginning. With the introduction of the concept of "smart earth", it immediately received a wide range of attention from the united states. And further establishes the position of the internet of things in network development. The internet of things is called as the third wave of the world information industry, represents the next generation of information development technology, and is regarded as the key technical field for dealing with the international financial crisis and the happy economy by all countries in the world. However, compared with the traditional internet, the number of terminals and the amount of transmitted information in the internet of things are larger, which brings a serious challenge to the QoS guarantee of the internet of things, and the design of the architecture of the internet of things is more important.
The cognitive technology refers to that transmission parameters are flexibly and dynamically changed on the basis of the cognition and the definition of a radio environment or a network state, so that the spectrum utilization efficiency is improved to the maximum extent. Currently, there are two specific applications of cognitive technology: cognitive radio and cognitive networks. Cognitive radio focuses on the cognitive capabilities of the radio, while cognitive networks focus on the application of cognitive functions in a single or multiple networks. Cognitive networks are able to perceive network conditions and make decisions, reasoning, learning, and take appropriate actions based thereon. In order to better manage and control the complexity of the network and improve the QoS of the network and the service experience of a user, the cognitive network introduces artificial intelligence related technology into the network, so that the network has the capabilities of self-management, self-learning and self-optimization, and the controllability, manageability and trustiness of the network are really realized.
However, there is only a few design related to the internet of things for small town cognition, and the existing internet of things architecture does not have self-management, self-learning and self-optimization cognitive capabilities, and only can change network configuration according to a preset threshold value.
Disclosure of Invention
The invention aims to provide a method for constructing a town cognition matrix Internet of things, which can realize the behavior cognition ability of the Internet of things constructed in a region to roles.
In order to achieve the purpose, the invention provides a method for constructing a town cognitive matrix internet of things, which comprises the following steps:
s1, carrying out location association construction on the composition elements in the region based on a knowledge graph;
s2, constructing a probe grid for detecting the mobile equipment in the area;
s3, performing matrix acquisition construction for acquiring the mobile data of the mobile equipment in the area, and performing data labeling construction for labeling the mobile data;
and S4, carrying out topology convolution construction for machine training and learning and carrying out prediction classification construction based on the mobile data.
According to an aspect of the present invention, the constituent elements include: a business activity zone, a server post zone and a client activity zone; step S1, including;
s11, carrying out location association on the distribution space of the commercial activity location and the shop business number based on the knowledge graph;
s12, carrying out location association on the server post location space and the server information in the business activity location based on the knowledge graph;
and S13, carrying out position association on the server post position space and the client movable position space based on the knowledge graph.
According to an aspect of the present invention, the step S2 includes:
s21, constructing a minimum triangular grid unit comprising three mobile equipment probes based on a topological structure;
s22, performing effective iteration based on the minimum triangular grid unit, and optimizing the minimum triangular grid unit according to an effective iteration result;
and S23, installing the mobile equipment probe based on the optimized vertex of the minimum triangular grid unit.
According to an aspect of the invention, in step S3, the step of performing a matrix acquisition construction for acquiring the movement data of the mobile device in the area includes:
s31, detecting the mobile equipment based on a single mobile equipment probe in the minimum triangular grid unit and associating the acquired mobile data;
s32, detecting the mobile equipment based on the three mobile equipment probes in the minimum triangular grid unit and calculating the associated mobile data, wherein in the process of calculating the associated mobile data, the three mobile equipment probes perform associated positioning calculation on three different signal intensities of the same mobile equipment;
s33, establishing position track data records of all roles in the area, wherein the position track data records are position set data records of the role association time, and the roles comprise waiters, customers, passerby and visitors.
According to an aspect of the present invention, in step S3, the step of constructing a data label for labeling the mobile data includes:
s34, performing attendant data annotation based on the attendant position track, wherein data records of the attendant position track are integrated into a data matrix or a data set associated with the attendant time position track;
s35, marking customer data based on the customer position track, wherein data records of the customer position track are integrated into a data matrix or a data set associated with the customer time position track;
s36, carrying out passenger data annotation based on the passenger position track, wherein data records of the passenger position track are integrated into a data matrix or a data set associated with the passenger time position track;
s37, carrying out passenger detection data annotation based on the passenger detection position track, wherein data records of the passenger detection position track are integrated into a data matrix or a data set associated with the passenger detection time position track.
According to an aspect of the present invention, in step S4, the step of performing topology convolution construction for machine training learning based on the movement data includes:
s41, performing data planning of machine learning training based on topological convolution, wherein the data planning is to convert data labels into machine learning training data in proportion through a label planning conversion interface, and the data labels are customer data labels, road passenger data labels, visitor data labels and waiter data labels respectively;
s42, training and learning of training data are carried out based on topological convolution, wherein the training and learning is to transfer the training data to GCN deep neural network machine learning based on a training input conversion interface, the GCN deep neural network machine learning is to output input topological matrix data to input data of a linear neural network through series of GCN topological convolution operations and a pooling conversion interface, and the series of GCN topological convolution operations are processes of outputting corresponding convolution operation data through convolution template operations;
s43, parameter adjusting learning of parameter adjusting data is carried out based on topological convolution, wherein the parameter adjusting learning is parameter optimization adjusting learning of the convolution template, the parameter optimization adjusting learning is convolution template parameter optimization adjusting learning based on back propagation gradient calculation, and input data of the convolution template parameter optimization adjusting learning is parameter adjusting data;
s44, test learning of test data is carried out based on topological convolution, wherein the test learning is the test learning of the output effect of the classification learning of the linear neural network added on the basis of the training learning and parameter adjusting learning steps, the classified data output effect test learning is to calculate tensor as input through a linear classification neural network into classified data output, the output value numbers of the classified data output are 1, 2 and 3, and the output value numbers 1, 2 and 3 represent the prediction classification of customers, passers-by and visitors of the test data respectively, the tensor is a tensor representation of the two-dimensional matrix eigen input data that is internally converted into a one-dimensional matrix eigen data set, the two-dimensional matrix characteristic input data is operation output data based on GCN deep neural network machine learning, and the input data of test learning is test data.
According to one aspect of the present invention, in step S41, the data planning is to scale the data labels into the machine learning training data through the label plan conversion interface by 70% training data, 15% parameter adjustment data, and 15% test data,
according to an aspect of the present invention, in step S4, the step of performing the predictive classification construction includes:
s45, acquiring activity data of all the mobile equipment in a minimum triangular grid unit based on real-time working data, wherein the real-time working data is the activity data of the mobile equipment in an unmarked area, and the activity data comprises the address data of a wireless network card of the mobile equipment and the signal intensity information data of the mobile equipment;
s46, performing real-time work data input and operation based on topological convolution operation, wherein the real-time work data input and operation is a process of inputting the real-time work data through a practice conversion interface and outputting the real-time work data through GCN deep neural network machine learning operation on the basis of the steps S42 to S44;
s47, classified data output is carried out based on linear classification neural network operation, namely input data are subjected to linear classification neural network operation and classified data output, the classified data output is that output value numbers are 1, 2 and 3, the output value numbers 1, 2 and 3 represent unmarked real-time working data classification, and the unmarked real-time working data classification is anonymous and unmarked anonymous customers, anonymous road customers and anonymous visitor prediction classification;
according to an aspect of the invention, further comprising:
s5, enabling labeling portrait technology based on behavior prediction;
and S6, enabling a network cooperation technology based on space-time metering.
According to an aspect of the present invention, the step S5 includes:
s51, behavior prediction data marking, wherein the behavior prediction data marking comprises shop detection behavior prediction data marking based on the conversion of a passerby into a visitor, cabinet detection behavior prediction data marking based on the conversion of the visitor into a customer, and purchasing behavior prediction data marking based on the conversion of an old customer into new shopping;
s52, behavior prediction learning training, wherein the training data planning comprises machine learning training data planning based on behavior prediction topological convolution, training learning based on behavior prediction topological convolution training data, parameter adjusting learning based on behavior prediction topological convolution parameter adjusting data, and test learning based on behavior prediction topological convolution test data;
and S53, labeling portrait classification output, wherein the labeling portrait classification output comprises the steps of collecting the mobile equipment in a minimum triangular grid unit based on labeling portrait real-time working data, inputting and calculating the labeling portrait topology convolution operation based on real-time working data, and outputting classification data based on labeling portrait linear classification neural network operation.
According to an aspect of the present invention, in step S51, in the step of labeling the forecast data based on the probe store behavior converted from the road passenger, the label of the forecast data of the probe store behavior is to label the forecast data of the probe store relationship affinity and sparsity locus description converted from the waiter and the road passenger, the description of the probe store relationship affinity and sparsity locus is to calculate the probe store difference value associated between the waiter position locus and the road passenger position locus, and the calculation of the probe store difference value is to grasp the calculation based on the probe store association trend characteristics of the matrix;
in the step of converting the visitor into the forecast data of the customer, the forecast data of the visitor is labeled, namely the forecast data of the visitor converted from the description of the visitor's visit cabinet relationship affinity and sparsity locus of the attendant and the visitor is labeled, the description of the visit cabinet relationship affinity and sparsity locus is the calculation of the visit cabinet difference value associated between the attendant's position locus and the visitor's position locus, and the calculation of the visit cabinet difference value locus is the calculation of the grabbing of the visit cabinet association trend characteristic based on the matrix;
in the step of converting old customers into purchasing behavior prediction data labels of new shopping, the purchasing behavior prediction data labels are obtained by converting purchasing relationship affinity and sparsity track descriptions of old customers and waiters into converted purchasing behavior prediction data labels, the purchasing relationship affinity and sparsity track descriptions are purchasing difference value calculation of the waiter position track and the old customer position track in association, and the purchasing difference value track calculation is purchasing association trend characteristic capture calculation based on a matrix.
According to an aspect of the present invention, in step S52, in the step of planning machine learning training data based on behavior prediction topological convolution, the planning of machine learning training data is to transfer data labels into machine learning training data in proportion through a label planning conversion interface, where the data labels are respectively store exploration behavior prediction data labels, probe cabinet behavior prediction data labels, purchase behavior prediction data labels, and waiter data labels;
in the step of training and learning based on behavior prediction topological convolution training data, the training and learning is to transfer the training data to GCN deep neural network machine learning based on a training input conversion interface, the GCN deep neural network machine learning is to output input topological matrix data to input data of a linear neural network through series of GCN topological convolution operations and a pooling conversion interface, and the series of GCN topological convolution operations are processes of outputting corresponding convolution operation data through convolution template operations;
a parameter adjusting learning step of predicting topological convolution parameter adjusting data based on behaviors, wherein the parameter adjusting learning is parameter optimization adjustment learning of the convolution template, the parameter optimization adjustment learning is convolution template parameter optimization adjustment learning based on back propagation gradient calculation, and input data of the convolution template parameter optimization adjustment learning is parameter adjusting data;
a test learning step based on the behavior prediction topological convolution test data, wherein the test learning is a classification data output effect test learning which adds linear neural network classification learning on the basis of the training learning and the parameter adjusting learning, the classified data output effect test learning is to calculate tensor as input through a linear classification neural network into classified data output, the output value numbers of the classified data output are 1, 2 and 3, and the output value numbers are 1, 2 and 3, which represent the classification of the store visit behavior prediction, the cabinet visit behavior prediction and the purchase behavior prediction of the test data respectively, the tensor is a tensor representation of the two-dimensional matrix eigen input data that is internally converted into a one-dimensional matrix eigen data set, the two-dimensional matrix characteristic input data is operation output data based on GCN deep neural network machine learning, and the input data of test learning is test data.
According to an aspect of the present invention, in step S53, in the step of collecting the mobile device in the minimum triangular grid cell based on the labeling portrait real-time working data, the labeling portrait real-time working data is the activity data of the mobile device in the unmarked and unmarked area, and the activity data includes the wireless network card address data of the mobile device and the signal strength information data of the mobile device;
in the step of inputting and calculating the real-time working data based on the labeling portrait topological convolution operation, the real-time working data is input through the practice conversion interface and output through the GCN deep neural network machine learning operation on the basis of the step S52;
the step of outputting classification data based on labeling portrait linear classification neural network operation is a process that input data is subjected to linear classification neural network operation and classification data output, wherein the classification data output is that output value numbers are 1, 2 and 3, the output value numbers 1, 2 and 3 are unlabeled real-time working data classifications, and the unlabeled real-time working data classifications are anonymous and unlabeled labeling anonymous customer, anonymous passerby and anonymous visitor labeling prediction classifications, and a shop detection behavior prediction portrait, a cabinet detection behavior prediction portrait and a purchase behavior prediction portrait.
According to an aspect of the present invention, the step S6 includes:
s61, performing space-time metering analysis visualization, wherein the space-time metering analysis visualization comprises time-space coordinate system design, panel data heat degree calculation presentation and time dimension data analysis visualization;
and S62, network cooperative scheduling of node service.
According to an aspect of the present invention, in step S61, the step of designing a time-space coordinate system is based on the time-space coordinate system design within the area, wherein the time-space coordinate system setting is to convert a three-dimensional area position coordinate system within the area into a two-dimensional plane area position coordinate system, and the two-dimensional plane area position coordinate system is to represent each floor area as a dimensional value of each area of the same two-dimensional plane coordinate system with different magnitude of values;
in the step of calculating and presenting panel data heat, calculating and presenting the panel data heat based on regional space-time measurement, wherein the regional space-time measurement is to arrange probe grid units of each place on each coordinate position in a two-dimensional plane coordinate system and measure and count frequency values of events of different people acquired by a probe according to time, and the panel data heat calculation presentation is to express the frequency values of calculation and statistics obtained on each time coordinate section as heat values, as heat colors and on the time coordinate section;
and time dimension data analysis visualization, wherein the time dimension data is data of a plotted curve acquired and calculated according to set time on the basis of a probe grid at set position coordinates, the time dimension data analysis is data change trend analysis of the plotted curve changing with time, and the data change trend analysis of the curve changing with time is visualized.
According to an aspect of the present invention, in the step of network collaborative scheduling of node services in step S62, node network collaboration is performed based on area space-time measurement, where the area space-time measurement is panel data heat analysis measurement, the node network is a node network composed of probe grid position nodes of an area space-time coordinate system, the node network collaboration is a network scheduling service for balancing node overheating, and the balancing node overheating is a network scheduling service for guiding excess number of people of the balancing node to other nodes for balanced offloading operation.
According to one scheme of the invention, the cognitive matrix Internet of things is constructed through zone bit association, so that the zone bit association of small-town shops, the zone bit association of server posts and the zone bit association of clients, passers and visitors of servers are realized; through the construction of the probe mesh, the construction of the minimum mesh unit of the WiFi probe of the three mobile phones is realized, the effective iteration of the minimum triangular mesh unit is realized, and the acquisition of the WiFi probe of the mobile phone at the vertex of the minimum triangular mesh unit is realized; through matrix acquisition construction, detection correlation acquisition records of a WiFi probe of a single-point mobile phone are realized, correlation calculation of data acquired by a WiFi probe of a three-vertex mobile phone with a minimum grid unit is realized, and position track data records of a client visitor of a waiter are realized; realizing attendant data annotation of attendant position tracks, realizing customer data annotation of customer position tracks, realizing passenger data annotation of passenger position tracks and realizing passenger detection data annotation of passenger detection position tracks by data annotation construction; through the topology convolution construction, the machine learning training data planning of the topology convolution is realized, the training learning of the topology convolution training data is realized, the parameter adjusting learning of the topology convolution parameter adjusting data is realized, and the test learning of the topology convolution test data is realized; through the prediction classification construction, the acquisition of the WiFi probe minimum matrix unit of the mobile phone of real-time working data is realized, the real-time working data input and operation of topological convolution operation is realized, and the classification data output step of linear classification neural network operation is realized. Therefore, the invention is constructed by the six technologies, and realizes the process from learning and understanding the time and position track knowledge of the mobile phone taking personnel in the area to predicting and recognizing the identities of customers, passers, visitors and waiters in the town (system).
According to one scheme of the invention, the cognitive matrix Internet of things enables by a labeling portrait technology based on behavior prediction, so that behavior prediction data labeling, behavior prediction learning training and behavior prediction labeling portrait classification output are realized; therefore, the invention realizes the labeling and portrait process of people for the small town (system) from learning and understanding the interactive relation change knowledge of customers, passerby, visitors and waiters in the area based on the behavior prediction to the behavior prediction cognition based on the shop detection, the cabinet detection and the purchase.
According to the scheme of the invention, the time-space metering analysis visualization and the network cooperative scheduling of the node service are realized through the energization of the time-space metering-based network cooperative technology. Therefore, the invention realizes the node network cooperative service capability based on the space-time metering analysis through the technology.
According to one scheme of the invention, a reverse position verification mobile phone is adopted, and an error reduction value is reversely transmitted to a position where a theoretical layout point is not overlapped with an actual layout point through N times of reverse verification measurement, so that the theoretical layout point and the actual layout point tend to be overlapped, the layout error of a matrix unit is eliminated, and the data acquisition accuracy of a probe in a minimum grid is improved.
Drawings
Fig. 1 schematically shows a connection structure diagram of the internet of things according to an embodiment of the present invention;
FIG. 2 schematically represents a GCN deep neural network and its machine learning architecture diagram, according to an embodiment of the present invention;
FIG. 3 is a schematic representation of a behavior prediction knowledge-graph and its labeling portraits technical relationship in accordance with an embodiment of the present invention;
FIG. 4 is a schematic representation of spatiotemporal metrics and node network synergy analysis diagram according to an embodiment of the present invention.
Detailed Description
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the embodiments will be briefly described below. It is obvious that the drawings in the following description are only some embodiments of the invention, and that for a person skilled in the art, other drawings can be derived from them without inventive effort.
In describing embodiments of the present invention, the terms "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," and the like are used in an orientation or positional relationship that is based on the orientation or positional relationship shown in the associated drawings, which is for convenience and simplicity of description only, and does not indicate or imply that the referenced device or element must have a particular orientation, be constructed and operated in a particular orientation, and thus, the above-described terms should not be construed as limiting the present invention.
The present invention is described in detail below with reference to the drawings and the specific embodiments, which are not repeated herein, but the embodiments of the present invention are not limited to the following embodiments.
According to an embodiment of the invention, the method for constructing the town cognitive matrix internet of things comprises the following steps:
s1, carrying out location association construction on the composition elements in the region based on a knowledge graph;
s2, constructing a probe grid of the mobile equipment in the area;
s3, performing matrix acquisition construction for acquiring mobile data of the mobile equipment in the region, and performing data labeling construction for labeling the mobile data;
and S4, carrying out topology convolution construction for machine training and learning based on the mobile data, and carrying out prediction classification construction.
According to one embodiment of the invention, the cognitive matrix internet of things is suitable for being operated in an area with a certain area, and in the embodiment, the area suitable for the cognitive matrix internet of things can be a small town, namely a characteristic small town, which is located in town and country areas and is a small area building space which integrates leisure travel, trade and trade logistics, …, education and technology, traditional culture, beautiful and pleasant and the like as characteristic service contents. Of course, the area of the present invention may also be in other forms, and will not be described herein. In the present embodiment, the constituent elements include: business activity zone, attendant post zone, customer activity zone. The location is the place occupied by a certain object, but also has the meaning in terms of position, layout, distribution, positional relationship, and the like. Since location theory is limited to studying human activities for survival and development, location is the place occupied by human activities (human behavior) in this sense.
In the present embodiment, step S1 includes;
s11, carrying out location association on the distribution space of the commercial activity location and the store business number based on a knowledge graph; the knowledge graph is an artificial intelligence NLP technical research object based on the relation knowledge category, focuses on researching the relations of people and people, people and objects and the like, is called as the technical theory of AI knowledge entity relation network, and becomes an effective tool for researching people-sum relation.
S12, carrying out zone bit association on the server post zone bit space and the server information in the business activity zone bit based on the knowledge graph;
and S13, carrying out position association on the server post position space and the client activity position space based on the knowledge graph.
According to an embodiment of the present invention, step S2 includes:
s21, constructing a minimum triangular grid unit comprising three mobile equipment probes based on the topological structure. In this embodiment, in the step of constructing the minimum triangular grid cell, the mobile device takes a mobile phone as an example to construct a three-phone WiFi probe minimum grid cell. In the embodiment, the three-mobile-phone WiFi probe minimum grid unit is constructed by respectively arranging three mobile-phone WiFi probes at three positions in a small town area space, wherein the three positions are arranged primarily at triangular edge distances between the probe positions and 20 meters and 50 meters, and the triangular edge distances are arranged primarily at angles at which any internal angle of the triangle is larger than zero.
And S22, performing effective iteration based on the minimum triangular grid unit, and optimizing the minimum triangular grid unit according to an effective iteration result. In this embodiment, the effective iteration is implemented by optimizing an optimal effect approximation of the coverage of the mesh cells by the test handset, the optimal effect approximation is implemented by forward and backward adjustment replacement of more optimal effect cells, the forward and backward adjustment replacement is an extended adjustment replacement of the forward and backward cells or an edge distance and angle adjustment replacement, the extended adjustment replacement is an extended coverage implementation of increasing the mesh, and the edge distance and angle adjustment replacement is an effective coverage implementation of scaling the edge distance and the angle to change the size and the shape of the mesh.
And S23, mounting a mobile equipment probe based on the optimized vertex of the minimum triangular grid unit. In the embodiment, the mobile device probe can be acquired through self-making or market purchasing, and the acquired mobile device probe has the function of detecting and acquiring the MAC address and the signal intensity information of the WiFi network card of the mobile device such as a mobile phone of nearby personnel, wherein the nearby personnel comprise waiters, customers, passerby and visitors. Note that the serviceman refers to a generic name of a small township serviceman, and the serviceman in a narrow sense refers to a serviceman in a small township shop. The customer refers to a person who enters the store to purchase goods or services. A passerby refers to a person who does not enter a store and does not purchase goods and services. A visitor refers to a person who enters a store but does not purchase goods and services.
According to an embodiment of the present invention, in step S3, the step of performing a matrix acquisition construction for acquiring movement data of a mobile device in an area includes:
s31, detecting the mobile equipment based on a single mobile equipment probe in the minimum triangular grid unit and associating the collected mobile data. In the embodiment, the detection associated acquisition record based on the WiFi probe of the single-point mobile phone is an associated acquisition record of the WiFi network card Mac address and the signal intensity data of mobile devices such as waiters, customers, passerby and visitor mobile phones.
And S32, detecting the mobile equipment based on three mobile equipment probes in the minimum triangular grid unit and calculating the associated mobile data, wherein in the process of calculating the associated mobile data, the three mobile equipment probes perform associated positioning calculation on three different signal intensities of the same mobile equipment. In the embodiment, based on the correlation calculation step of the data acquired by the three-vertex position mobile phone WiFi probe of the minimum grid unit, the correlation calculation is that the three-vertex position mobile phone WiFi probe respectively performs correlation positioning calculation on three different signal intensities of the same Mac, and the correlation positioning calculation is position track determination calculation that three different signal intensity ratios are correlated with the three-vertex arrangement edge distance lengths of the probe.
And S33, establishing position track data records of all roles in the area, wherein the position track data records are position set data records of role association time, and the roles comprise waiters, customers, passers-by and visitors. In the present embodiment, the position is a position at which the attendant relates time to the customer, the road passenger, and the probe based on the position trajectory data records of the attendant, the customer, the road passenger, and the probe, and the trajectory data records are position aggregate data records at which the attendant, the customer, the road passenger, and the probe relate time.
According to an embodiment of the present invention, the step of constructing the data annotation for annotating the mobile data in step S3 includes:
s34, performing attendant data annotation based on the attendant position track, wherein in the step, data records of the attendant position track are integrated into a data matrix or a data set associated with the attendant time position track;
s35, marking customer data based on the customer position track, wherein in the step, data records of the customer position track are integrated into a data matrix or a data set associated with the customer time position track;
s36, carrying out passenger data annotation based on the passenger position track, wherein in the step, data records of the passenger position track are integrated into a data matrix or a data set associated with the passenger time position track;
and S37, marking the visitor data based on the visitor position track, wherein in the step, the data records of the visitor position track are integrated into a data matrix or a data set associated with the visitor time position track.
According to an embodiment of the present invention, in step S4, the step of performing topology convolution construction for machine training learning based on movement data includes:
and S41, planning data of machine learning training based on topological convolution. In the embodiment, the topological convolution is based on the topological graph theory convolution of GCN deep neural network machine learning, the data planning is to transfer data labels into machine learning training data through a label planning conversion interface in proportion, and in the process of transferring the data labels into the machine learning training data in proportion, the data labels are transferred through the conversion interface in proportion according to the proportion of 70% of training data, 15% of parameter adjustment data and 15% of test data, and the data labels are customer data labels, road passenger data labels, visitor data labels and attendant data labels respectively.
And S42, training and learning of training data based on topological convolution. In the embodiment, the training learning is to transfer the training data to the GCN-based deep neural network machine learning through the training input conversion interface, and the codes of the GCN deep neural network machine learning can adopt the existing GCN deep neural network codes. In the embodiment, the GCN deep neural network machine learning is to output input topology matrix data as input data of a linear neural network through a series of GCN topology convolution operations and a pooling conversion interface. In the present embodiment, the series of GCN topology convolution operations includes a first-level convolution operation, a second-level convolution operation, a third-level convolution operation, …, and an N-level convolution operation, wherein, the first-stage convolution operation is a process of inputting topology matrix data and outputting first-stage convolution operation data through first-stage convolution template operation, the second-stage convolution operation is a process of outputting second-stage convolution operation data through second-stage convolution template operation by taking the first-stage convolution operation data as input, the third-stage convolution operation is a process of outputting third-stage convolution operation data through third-stage convolution template operation by taking the second-stage convolution operation data as input, …, the N-stage convolution operation is a process of outputting N-stage convolution operation data through N-stage convolution template operation by taking N-1-stage convolution operation data as input, the N-level numerical value of the template and the quantity M of the templates at each level are finally determined by the comprehensive balance of the learning training process, the effect and the efficiency. In the present embodiment, the input data used for training learning in the present step is training data.
And S43, performing parameter adjustment learning of parameter adjustment data based on topological convolution. In the present embodiment, the parameter tuning learning in this step is parameter optimization tuning learning of each convolution template M of the first, second, third, …, and N stages in the aforementioned step, the parameter optimization tuning learning is convolution template parameter optimization tuning learning based on back propagation gradient calculation, and the input data of the parameter tuning learning is the parameter tuning data.
And S44, testing and learning of the test data based on the topological convolution. In this embodiment, the test learning in this step is a classification data output effect test learning in which linear neural network classification learning is added on the basis of the training learning and parameter-adjusting learning step in the preceding step, the classification data output effect test learning is to calculate tensors as input linear classification neural networks into classification data output, the output value numbers of the classification data output are 1, 2, and 3, and the output value numbers 1, 2, and 3 respectively represent prediction classifications of customers, road passengers, and probes in test data, the tensors are tensor representations in which two-dimensional matrix feature input data are internally converted into one-dimensional matrix feature data sets, the input data are operation output data based on GCN deep neural network machine learning, and the input data of the test learning are test data.
According to an embodiment of the present invention, the step of constructing the prediction classification in step S4 includes:
and S45, acquiring the activity data of all the mobile devices in the minimum triangular grid unit based on the real-time working data. In this embodiment, the real-time working data in this step is unlabelled small-town daily live mobile phone activity data, the mobile phone activity data is mobile phone WiFi network card Mac address and intensity information data thereof collected by the minimum matrix unit of the mobile phone WiFi probe during small-town daily working, and the mobile phone is a mobile phone held by small-town store employees, anonymous customers, anonymous passerby and anonymous visitor.
And S46, performing real-time working data input and operation based on topological convolution operation. In this embodiment, the real-time work data input and operation in this step is a process of inputting the real-time work data through the training conversion interface and outputting the real-time work data through the GCN deep neural network machine learning operation based on the training learning, parameter adjusting learning and testing learning steps in the above steps.
And S47, outputting classified data based on linear classification neural network operation. In this embodiment, in the step, the classification data output of the linear classification neural network operation, i.e., the process of calculating the input data as the classification data output through the linear classification neural network operation, the classification data output is that the output value numbers are 1, 2, and 3, and the output value numbers 1, 2, and 3 respectively represent the classification of the unlabeled real-time working data, and the unlabeled real-time working data classification is anonymous and unlabeled anonymous customer, anonymous road passenger, and anonymous visitor prediction classification;
according to an embodiment of the invention, the method for constructing the town cognitive matrix internet of things further comprises the following steps:
s5, enabling labeling portrait technology based on behavior prediction;
and S6, enabling a network cooperation technology based on space-time metering.
According to an embodiment of the present invention, step S5 includes:
and S51, behavior prediction data marking, wherein the behavior prediction data marking comprises shop detection behavior prediction data marking based on the conversion of a passerby into a visitor, cabinet detection behavior prediction data marking based on the conversion of the visitor into a customer, and purchasing behavior prediction data marking based on the conversion of an old customer into a new shopping.
In the embodiment, in the step of labeling the detection shop behavior prediction data based on the conversion from the road passenger to the visitor, the detection shop behavior prediction data labeling is to label the detection shop behavior prediction data converted from the detection shop relationship affinity and sparsity track description of the waiter and the road passenger, the detection shop relationship affinity and sparsity track description is the detection shop difference calculation of the correlation between the waiter position track and the road passenger position track, and the detection shop difference calculation is the detection shop correlation trend feature capture calculation based on the matrix;
in the step of converting the visitor into the forecast data of the customer, the forecast data of the visitor is labeled, namely the forecast data of the visitor converted from the description of the relation between the waiter and the visitor is labeled, the description of the relation between the waiter and the visitor is the difference calculation of the visit cabinet related to the position track of the waiter and the position track of the visitor, and the calculation of the difference track of the visit cabinet is the grabbing calculation of the relation trend characteristic of the visit cabinet based on the matrix;
in the step of converting old customers into purchasing behavior prediction data labels of new shopping, the purchasing behavior prediction data labels are converted from purchasing relationship affinity and sparsity track descriptions of old customers and waiters, the purchasing relationship affinity and sparsity track descriptions are purchasing difference calculation of the relationship between the waiter position track and the old customer position track, and the purchasing difference track calculation is purchasing association trend characteristic capture calculation based on a matrix.
S52, behavior prediction learning training, wherein the training data planning based on the behavior prediction topological convolution, the training learning based on the behavior prediction topological convolution training data, the parameter adjusting learning based on the behavior prediction topological convolution parameter adjusting data and the testing learning based on the behavior prediction topological convolution testing data are included.
In this embodiment, the step of planning the machine learning training data based on the behavior prediction topological convolution includes a step of planning the machine learning training data based on the topological convolution, where the topological convolution is a topological graph theory convolution based on GCN deep neural network machine learning, the data planning is to transfer data labels into the machine learning training data through a label planning conversion interface in proportion, the transfer is to transfer the data labels into the machine learning training data through a label planning conversion interface in proportion according to the proportion of 70% of training data, 15% of parameter adjustment data and 15% of test data, and the data labels are respectively store exploration behavior prediction data labels, cabinet exploration behavior prediction data labels, purchase behavior prediction data labels and server data labels.
The training and learning step based on the topological convolution training data of behavior prediction comprises the training and learning step based on the topological convolution training data, the training and learning step is to transfer the training data to GCN deep neural network machine learning through a training input conversion interface, codes of the GCN deep neural network machine learning can adopt the existing GCN codes, the machine learning step is to output input topological matrix data to input data of a linear neural network through a series of GCN topological convolution operations and a pooling conversion interface, the series of GCN topological convolution operations comprise a first-level convolution operation, a second-level convolution operation, a third-level convolution operation, … and an N-level convolution operation, wherein the first-level convolution operation is a process of outputting first-level convolution operation data through a first-level convolution template operation on the input topological matrix data, and the second-level convolution operation is a process of outputting second-level convolution operation data through a second-level convolution template operation on the input topological matrix data, the three-level convolution operation is a process of outputting three-level convolution operation data through three-level convolution template operation by taking the two-level convolution operation data as input, …, the N-level convolution operation is a process of outputting N-level convolution operation data through N-level convolution template operation by taking the N-1-level convolution operation data as input, and the N-level numerical value of the convolution template and the number M of each level of templates are finally determined through the comprehensive balance of the learning training process, the effect and the efficiency. In the present embodiment, input data for training learning is training data.
The step of tuning learning based on the behavior prediction topological convolution tuning parameter data comprises the tuning learning based on the topological convolution tuning parameter data, wherein the tuning learning is parameter optimization adjustment learning of convolution templates M of the first level, the second level, the third level, the … level and the N level, and the parameter optimization adjustment learning is convolution template parameter optimization adjustment learning based on back propagation gradient calculation. In the present embodiment, the input data for the parameter adjustment learning is parameter adjustment data.
The step of test learning based on the behavior prediction topological convolution test data comprises a test learning step based on the topological convolution test data, the test learning is the classified data output effect test learning of adding linear neural network classification learning on the basis of the training learning and parameter adjusting learning step of the previous step, the classified data output effect test learning is to calculate tensor as input data to be classified data output by a linear classification neural network, the output value numbers of the classified data output are 1, 2 and 3, the output value numbers 1, 2 and 3 respectively represent the classification of the shop detection behavior prediction, the cabinet detection behavior prediction and the purchase behavior prediction of the test data, the tensor is the tensor representation of two-dimensional matrix characteristic input data which is internally converted into a one-dimensional matrix characteristic data set, and the input data is the calculated output data based on the GCN deep neural network machine learning, the input data for the test learning is test data.
And S53, labeling portrait classification output, wherein the labeling portrait classification output comprises the steps of collecting the mobile equipment in a minimum triangular grid unit based on labeling portrait real-time working data, inputting and calculating the labeling portrait topology convolution operation based on real-time working data, and outputting classification data based on labeling portrait linear classification neural network operation.
In this embodiment, the acquisition is performed by the minimum constructed triangular mesh unit. In this step, the mobile phone WiFi probe minimum matrix unit based on real-time working data is collected, the real-time working data is the mobile phone activity data of the small-birthday site which is not marked and does not need to be marked, and the mobile phone activity data is the mobile phone WiFi network card Mac address and the intensity information data thereof collected by the mobile phone WiFi probe minimum matrix unit during the small-birthday normal work. In the present embodiment, the mobile phone used for collecting data is a mobile phone held by a small-town store employee, an anonymous customer, an anonymous passerby, and an anonymous seeker.
In the embodiment, the step of inputting and calculating the real-time working data based on the labeling portrait topological convolution operation includes the step of inputting and calculating the real-time working data based on the topological convolution operation, and the step of inputting and calculating the real-time working data is a process of inputting the real-time working data through the training input conversion interface and outputting the real-time working data through the GCN deep neural network machine learning operation on the basis of the steps of training learning, parameter adjusting learning and testing learning in the previous step.
In this embodiment, the step of outputting the classification data calculated by the linear classification neural network based on the labeling portrait includes the step of outputting the classification data calculated by the linear classification neural network, wherein the output of the classification data calculated by the linear classification neural network is a process of outputting the input data calculated by the linear classification neural network as the classification data, the output data is output values 1, 2 and 3, and the output values 1, 2 and 3 respectively represent the classification of the unlabeled real-time working data, and the classification of the unlabeled real-time working data is the labeling prediction classification and the probing behavior prediction portrait of the anonymous and unlabeled anonymous customer, anonymous passerby and anonymous probing customer, the probing behavior prediction portrait and the purchasing behavior prediction portrait.
According to an embodiment of the present invention, step S6 includes:
and S61, performing space-time metrological analysis visualization, wherein the space-time metrological analysis visualization comprises time-space coordinate system design, panel data heat degree calculation presentation and time dimension data analysis visualization.
In this embodiment, in the step of designing the time-space coordinate system, the time-space coordinate system is designed based on a time-space coordinate system in the area, where the setting of the time-space coordinate system is to convert a three-dimensional area position coordinate system in the area into a two-dimensional plane area position coordinate system, and the two-dimensional plane area position coordinate system is to represent each floor area as a dimensional value of each area in the same two-dimensional plane coordinate system with different magnitude of values.
In this embodiment, in the step of presenting the panel data heat calculation, the panel data heat calculation is presented based on a region space-time measurement, where the region space-time measurement is to set the probe grid unit of each location at each coordinate position in the two-dimensional plane coordinate system and to perform measurement statistics on frequency values of events occurring to different people collected by the probe according to time, and the panel data heat calculation presentation is to represent the frequency values of the calculation statistics obtained on each time coordinate section as heat values, as heat colors, and as expressed on the time coordinate section.
In the embodiment, the time dimension data analysis visualization is performed on the basis of regional space-time measurement, wherein the time dimension data is data of a plotted curve acquired and calculated according to set time on the basis of coordinates of a probe grid at a set position, the time dimension data analysis is data change trend analysis of the plotted curve changing along with time, and the data change trend analysis of the curve changing along with time is visually presented.
And S62, network cooperative scheduling of node service. In this embodiment, in the step of network cooperative scheduling of the node service, node network cooperation is performed based on area space-time measurement, the area space-time measurement is panel data heat analysis measurement, the node network is a node network composed of probe grid position nodes of an area space-time coordinate system, the node network cooperation is a network scheduling service for balancing node overheating, and the balancing node overheating is a network scheduling service for guiding the excessive number of people of the balancing node to other nodes for balanced shunting operation.
According to the method, the zone bit association of the small-town shop is realized, the station bit association of a server post is realized, and the zone bit association of a server guest is realized through the zone bit association construction; through the construction of the probe mesh, the construction of the minimum mesh unit of the WiFi probe of the three mobile phones is realized, the effective iteration of the minimum triangular mesh unit is realized, and the acquisition of the WiFi probe of the mobile phone at the vertex of the minimum triangular mesh unit is realized; through matrix acquisition construction, detection correlation acquisition records of a WiFi probe of a single-point mobile phone are realized, correlation calculation of data acquired by a WiFi probe of a three-vertex mobile phone with a minimum grid unit is realized, and position track data records of a client visitor of a waiter are realized; realizing attendant data annotation of attendant position tracks, realizing customer data annotation of customer position tracks, realizing passenger data annotation of passenger position tracks and realizing passenger detection data annotation of passenger detection position tracks by data annotation construction; through the topology convolution construction, the machine learning training data planning of the topology convolution is realized, the training learning of the topology convolution training data is realized, the parameter adjusting learning of the topology convolution parameter adjusting data is realized, and the test learning of the topology convolution test data is realized; through the prediction classification construction, the acquisition of the WiFi probe minimum matrix unit of the mobile phone of real-time working data is realized, the real-time working data input and operation of topological convolution operation is realized, and the classification data output step of linear classification neural network operation is realized. Therefore, the invention realizes the process from learning and understanding the time and position track knowledge of the mobile phone taking personnel in the area to predicting and recognizing the identities of customers, passers, visitors and waiters by the town (system) through the construction method.
According to the method, labeling portrait technology based on behavior prediction is enabled in the enabling technical characteristics, so that behavior prediction data labeling, behavior prediction learning training and behavior prediction labeling portrait classification output are realized; and the visualization of the time-space metering analysis and the network cooperative scheduling of the node service are realized through the energization of a time-space metering-based network cooperative technology. Therefore, the system realizes the process of labeling and portraying personnel from learning and understanding the interactive relation change knowledge based on behavior prediction of customers, passerby, visitors and waiters in the area to learning and understanding the processes of detecting shops, exploring cabinets and purchasing personnel based on behavior prediction cognition; and further, the node network based on the space-time metering analysis is cooperated with the service capability.
The construction method of the present invention will be described below by way of example with reference to the accompanying drawings.
As shown in fig. 1, according to the construction method of the present invention, a cognitive matrix internet of things in a region (e.g., a town) is constructed, wherein the construction method includes location association construction, probe grid construction, matrix acquisition construction, data annotation construction, topology convolution construction, and prediction classification construction.
As shown in fig. 1, in the present embodiment, a location association is constructed, and in the step of associating a small town store location based on a knowledge graph in the above-described step, it is preferable that the small town store location association is an association between a small town location distribution space 1 and a store number 2. In the knowledge-map based attendant position association step, the present invention preferably associates the attendant position association with the store position space 3 and the name or number of attendant 4. In the knowledge-map based attendant guest zone association step, the present invention preferably provides that the attendant guest zone association is an association of a store post zone space 3 and a guest zone space 5.
As shown in fig. 1, a probe grid construction is performed according to one embodiment of the present invention. In the embodiment, in the step of constructing the three-cell-phone WiFi probe 6 minimum grid unit 13 based on the probe grid, the three-cell-phone WiFi probe 6 minimum grid unit 13 is constructed by respectively arranging three cell-phone WiFi probes 6 at three positions in the small town area space 1, the three positions are arranged by primarily arranging the triangular edge distance 7 between the probe position distance 20 m and 50 m, and the primarily arranging the triangular edge distance 7 is arranged by arranging any angle of the triangle, wherein any inner angle of the triangle is larger than zero degree.
In this embodiment, in the effective iteration step based on the minimum triangular mesh unit 13, the effective iteration is implemented by the optimal effect approximation that the test handset optimizes the coverage of the mesh unit 13, the optimal effect approximation is implemented by the front and back adjustment replacement of the more optimal effect unit 13, the front and back adjustment replacement is the extension adjustment replacement of the front and back unit 13 or the edge 7 and angle adjustment replacement, the extension adjustment replacement is implemented by increasing the expanded coverage of the mesh, and the edge 7 and angle adjustment replacement are implemented by scaling the edge 7 and angle to change the size and shape of the mesh 13.
In the embodiment, based on the step of obtaining the mobile phone WiFi probe 6 at the vertex of the minimum triangular mesh unit 13, the mobile phone WiFi probe 6 may be made by self or purchased in the market, and it is preferable that the mobile phone WiFi probe 6 is purchased in the market in the present invention, and the purchased mobile phone WiFi probe 6 has a function of detecting and acquiring the mobile phone WiFi network card MAC address and the signal intensity information of the nearby waiters 4, the customers 10, the passerby 11, and the visitor 12.
As shown in fig. 1, according to one embodiment of the present invention, a matrix acquisition build is performed. In this embodiment, in the step of acquiring the association record of mobile phone detection based on the single-point WiFi probe 6, the association record is the association record of the mobile phone WiFi network card Mac address and the signal intensity data of the attendant 4, the customer 10, the passerby 11, and the visitor 12.
In the embodiment, in the correlation calculation step of the data collected by the three-vertex mobile phone WiFi probes 6 based on the minimum grid unit, the correlation calculation is that the three-vertex mobile phone WiFi probes 6 perform correlation positioning calculation on the same Mac with three different signal intensities, and the correlation positioning calculation is that the three different signal intensity ratios are correlated with the position track determination calculation of the length of the three-vertex arrangement edge distance 7 of the probes.
In the present embodiment, in the track 9 data recording step, the position of the attendant 4, the customer 10, the road passenger 11, and the probe 12 is a position at which the attendant 4 is associated with the time of the customer 10, the road passenger 11, and the probe 12, and the track 9 data record is a position set data record at which the attendant 4, the customer 10, the road passenger 11, and the probe 12 are associated with the time.
As shown in FIG. 1, according to one embodiment of the present invention, data annotation construction is performed. In the present embodiment, in the step of attendant 4 data annotation based on attendant 4 position trajectory 9, the attendant 4 data annotation is the integration of data records of attendant 4 position trajectory 9 into a data matrix or data set associated with attendant 4 time position trajectory.
In the present embodiment, in the step of marking the customer 10 data based on the customer 10 position trajectory 9, the customer 10 data marking is a step of integrating the data records of the customer 10 position trajectory 9 into a data matrix or a data set associated with the customer 10 time position trajectory.
In the present embodiment, in the step of the data annotation of the passenger 11 based on the position trajectory 9 of the passenger 11, the data annotation of the passenger 11 is a data matrix or a data set in which the data records of the position trajectory 9 of the passenger 11 are integrated into a time position trajectory of the passenger 11.
In this embodiment, in the step of data annotation of the probe 12 based on the position track 9 of the probe 12, the data annotation of the probe 12 is to integrate the data records of the position track 9 of the probe 12 into a data matrix or a data set associated with the time position track of the probe 12.
Referring to fig. 1 and 2, a topological convolution is constructed according to an embodiment of the present invention. In the present embodiment, in the step of planning the machine learning training data based on the topological convolution, the topological convolution is based on the topological graph theory convolution of GCN deep neural network machine learning (see A), the data planning is to scale the data labels (see C) into the machine learning training data (see B) through the label planning conversion interface (see D1), the scaling is achieved by scaling the data labels (see C) via the label plan conversion interface (see D1) (see B) by 70% training data (see B1), 15% parameter adjustment data (see B2), 15% test data (see B3), the data labels (see C) are a customer data label (see C1), a road passenger data label (see C2), a probe data label (see C3), and an attendant data label (see C4), respectively.
In the present embodiment, in the step of training learning of training data (see B1) based on topological convolution, the training learning is to transfer the training data (see B1) to GCN deep neural network-based machine learning (see a) through a training input conversion interface (see D2), the code of which is to adopt the GCN code of the related art, the machine learning (see a) is to output the input topological matrix data (see a1) to input data (see E2) of a linear neural network (see E) through a series of GCN topological convolution operations, which are a primary convolution operation, a secondary convolution operation, a tertiary convolution operation, …, an N-level convolution operation, the primary convolution operation is a process in which the input topological matrix data (see a1) is operated to output primary operation data (see A3) through a primary convolution template (see a2), the second-order convolution operation is a process of outputting the second-order convolution operation data (a5) by operating the second-order convolution template (see a4) with the first-order convolution operation data (see A3) as an input, and the third-order convolution operation data (see a7) by operating the third-order convolution template (see a6) with the second-order convolution operation data (see a5) as an input, and the N-order convolution operation data (see a9) is a process of outputting the N-order convolution operation data (see E2) by operating the N-order convolution template (see a10) with the N-order value of the convolution template and the template number M finally determined by learning the comprehensive balance of the training process, the effect, and the efficiency, and the input data to the training learning is the training data (see B1).
In the present embodiment, in the step of performing parameter tuning learning of parameter tuning data based on topological convolution, the parameter tuning learning is first, second, third, …, N-level convolution templates (see a2, a4, a6, a10) parameter optimization tuning learning, the parameter optimization tuning learning is convolution template parameter optimization tuning learning based on back propagation gradient calculation, and input data of the parameter tuning learning is parameter tuning data (see B2).
In the present embodiment, the test learning step based on the topological convolution test data is a classification data output (see E5) effect test learning of adding linear neural network (see E) classification learning to the training and parameter-adjusting learning step, the classification data output (see E5) effect test learning is to calculate tensor (see E3) as input via linear classification neural network (see E4) as classification data output (see E5), the output value numbers of the classification data output (see E5) are 1, 2 and 3, the output value numbers 1, 2 and 3 are respectively prediction classifications of customer 10, visitor 11 and visitor 12 of the test data, the tensor (see E3) is a tensor (see E3) representation in which two-dimensional matrix eigen input data (see E2) is internally converted into a matrix eigen data set, and the input data (see E2) is an arithmetic output (see a) of GCN deep neural network machine learning (see a) Data, input data of the test learning is test data (see B3).
Referring to fig. 1 and 2, according to an embodiment of the present invention, a prediction classification is constructed. In this embodiment, in the step of acquiring the minimum matrix unit of the mobile phone WiFi probe 6 based on the real-time working data, the real-time working data is the small-town daily field mobile phone activity data which is not labeled and does not need to be labeled, the mobile phone activity data is the mobile phone WiFi network card Mac address and the intensity information data thereof acquired by the minimum matrix unit 13 of the mobile phone WiFi probe 6 during the small-town daily working, and the mobile phone is the mobile phone held by the small-town store employee 6, the anonymous customer 10, the anonymous passerby 11, and the anonymous visitor 12.
In the present embodiment, in the step of inputting and computing the real-time working data based on the topological convolution operation, the real-time working data is input through the training input conversion interface (see D2) and output through the GCN deep neural network machine learning (see a) operation (see E2) based on the training, parameter adjustment, and test learning steps.
In the present embodiment, in the step of outputting (see E5) the classification data calculated based on the linear classification neural network (see E4), the classification data output (see E5) calculated by the linear classification neural network (see E4), i.e., the process of calculating the input data (see E2) as the classification data output (see E5) through the linear classification neural network (see E4), the classification data output (see E5) is that the output value numbers are 1, 2, and 3, and the output value numbers 1, 2, and 3 respectively represent the unlabeled real-time working data classification, i.e., the anonymous and unlabeled customer 10, anonymous passer 11, and anonymous snooper 12 predict the classification.
With reference to fig. 1 and 2, the cognitive matrix internet of things of the present invention is constructed by zone-location association, so as to realize zone-location association of a town 1 store 2, realize zone-location association of a server 4 at post 3, and realize zone-location association of a server 4 customer 10, a roadside passenger 11, and a visitor 12; through the construction of the probe grids, the construction of the minimum grid unit 13 of the WiFi probe 6 of the three mobile phones is realized, the effective iteration of the minimum triangular grid unit 13 is realized, and the acquisition of the WiFi probe 6 of the mobile phone at the vertex of the minimum triangular grid unit 13 is realized; through matrix acquisition construction, detection correlation acquisition records of the WiFi probe 6 of the single-point mobile phone are realized, correlation calculation of data acquired by the WiFi probe 6 of the three-vertex mobile phone with the minimum grid unit 13 is realized, and data records of position tracks 8 and 9 of a waiter customer 10, a passerby 11 and a visitor 12 are realized; through data annotation construction, realizing attendant data annotation (see C4) of attendant 4 position tracks 8 and 9, realizing customer 10 data annotation (see C1) of customer 10 position tracks 8 and 9, realizing visitor 11 data annotation (see C2) of visitor 11 position tracks 8 and 9, and realizing visitor 12 data annotation (see C3) of visitor 12 position tracks 8 and 9; through topology convolution construction, machine learning training data planning of topology convolution (see A), training learning of topology convolution training data (see B1), parameter tuning learning of topology convolution parameter tuning data (see B2) and test learning of topology convolution test data (see B3) are achieved; through the prediction classification construction, the acquisition of the mobile phone WiFi probe 6 minimum matrix unit 13 of real-time working data is realized, the real-time working data input (see D2) and operation of topological convolution operation is realized, and the classification data output step of linear classification neural network operation is realized (see E5). Therefore, the invention realizes the process of learning, understanding the knowledge of the time, position and track 8 and 9 of the mobile phone personnel 4, 10, 11 and 12 in the area and predicting and recognizing the identity of the customer 10, the road passenger 11, the visitor 12 and the waiter 4 by the construction of the six technologies in the town (system).
With reference to the building method shown in fig. 1, 3 and 4, the internet of things of cognitive matrixes in an area (such as a town) is built, and the method further comprises the step of enabling a labeling portrait technology based on behavior prediction, wherein the labeling portrait technology based on behavior prediction is enabled.
According to one embodiment of the invention, the step of labeling portrait technology enabling based on behavior prediction comprises behavior prediction data labeling, behavior prediction learning training and labeling portrait classification output.
In the process of labeling the behavior prediction data, the step of labeling the behavior prediction data based on the detection (see edge I) of the visitor 12 converted from the visitor 11 is included, the detection behavior (see edge I) prediction data labeling describes the converted detection (see edge I) behavior prediction data labeling by describing affinity and sparseness tracks 8 and 9 of the relation between the waiter 4 and the visitor 11, the detection (see edge I) behavior affinity and sparseness tracks 8 and 9 of the waiter 4 and the visitor 11 are described as the detection difference calculation of the relation between the waiter 4 position track 8 and the visitor 11 position track 9, and the detection (see edge I) difference track calculation is the capturing calculation of the relation trend characteristics of the matrix-based detection (see edge I).
In this embodiment, in the step of converting the probe 12 into the probe cabinet (see edge ii) behavior prediction data label of the customer 10, the probe cabinet (see edge ii) behavior prediction data label is a label of converted probe cabinet (see edge ii) behavior prediction data describing the probe cabinet (see edge ii) relationship affinity and sparseness locus 9 of the attendant 4 and the probe 12, the probe cabinet (see edge ii) relationship affinity and sparseness locus 9 of the attendant 4 and the probe 12 is a difference calculation of probe cabinets (see edge ii) associated with the position locus 8 of the attendant 4 and the position locus 9 of the probe 12, and the probe cabinet (see edge ii) difference locus calculation is a capturing calculation of associated trend characteristics of probe cabinets (see edge ii) based on a matrix.
In the present embodiment, the step of labeling purchase (see edge iii) behavior prediction data, which is converted into new shopping by old customer 10, is based on the step of labeling purchase (see edge iii) behavior prediction data, which is converted from the purchase (see edge iii) relationship affinity track 9 of old customer 10 and waiter 4, the purchase (see edge iii) relationship affinity track 9 of waiter 4 and old customer 10 describing the purchase (see edge iii) difference calculation of waiter 4 position track 8 and old customer 10 position track 9, and the purchase (see edge iii) difference track calculation being the purchase (see edge iii) association tendency feature grasping calculation based on the matrix.
In the process of behavior prediction learning training, as shown by combining fig. 3 and fig. 2, a step of planning machine learning training data based on behavior prediction topological convolution, which can refer to the content in the preceding steps, includes the step of planning machine learning training data based on topological convolution, which is topological graph theory convolution based on GCN deep neural network machine learning (see a), the data planning is to scale data labels (see C) into the machine learning training data (see B) through a label planning conversion interface (see D1), the scaling is to scale data labels (see C) into the machine learning training data (see B1), 15% parameter adjustment data (see B2), and 15% test data (see B3) into the conversion interface (see D1) through training and to scale data labels (see B), the data labels (see C) are store (see edge i) behavior prediction data label (see C1), explorer (see edge ii) behavior prediction data label (see C2), purchase (see edge iii) behavior prediction data label (see C3), and attendant data label (see C4), respectively.
In the present embodiment, the training learning step based on the behavior prediction topological convolution training data (see B1) is included, the learning training is referred to in the preceding steps, and includes the training learning step based on the topological convolution training data (see B1), the training learning step is to transfer the training data (see B1) to the GCN deep neural network machine learning (see a) through the training input conversion interface (see D2), the code of the GCN deep neural network machine learning (see a) can adopt the GCN code in the prior art, the machine learning (see a) is to output the input topological matrix data (see a1) to the input data (see E2) of the linear neural network (see E) through the series of GCN topological convolution operations, the series of GCN topological convolution operations are a first-order convolution operation, a second-order convolution operation, and a third-order operation, and the third-order operation is performed through the pooling conversion interface (see D3), Three-stage convolution operation …, N-stage convolution operation, the one-stage convolution operation being a process of outputting one-stage convolution operation data (see A3) by a one-stage convolution template (see a2) operation on input topology matrix data (see a1), the two-stage convolution operation being a process of outputting two-stage convolution operation data (see a5) by a two-stage convolution template (see a4) operation on input topology matrix data (see A3), the three-stage convolution operation being a process of outputting three-stage convolution operation data (see a7) by a three-stage convolution template (see A6) operation on input two-stage convolution operation data (see a5), the N-stage convolution operation data (see a9) being a process of outputting N-stage convolution operation data (see E2) by an N-stage convolution template (see a10) operation on input N-1-stage convolution operation data (see a9), the N-stage number of N-stage template and the number M of stages being trained by a learning process, The comprehensive balance of effect and efficiency is finally determined, and the input data of the training learning is training data (see B1).
In the present embodiment, the parameter tuning learning step based on the behavior prediction topological convolution parameter tuning data, which refers to the content in the preceding steps, includes a parameter tuning learning step based on the topological convolution parameter tuning data, which is parameter optimization tuning learning of convolution templates of the first, second, third, …, N levels (see a2, a4, a6, a10), which is convolution template parameter optimization tuning learning based on back propagation gradient calculation, and the input data of the parameter tuning learning is parameter tuning data (see B2);
in the present embodiment, the test learning step based on the behavior prediction topological convolution test data, which refers to the content of the preceding step, includes the test learning step based on the topological convolution test data, which is to add the classification data output (see E5) of the linear neural network (see E) classification learning to the training and parameter-adjusting learning step, the classification data output (see E5) effect test learning is to calculate the tensor (see E3) as the input to the linear classification neural network E4 as the classification data output (see E5), the output value numbers of the classification data output (see E5) are 1, 2, and 3, and the output value numbers 1, 2, and 3 respectively represent the classification of the shop behavior prediction, and the purchase behavior prediction of the test data, and the tensor (see E3) is the two-dimensional matrix eigen input data (see E2) which is internally converted into the one-dimensional matrix eigen data The tensor of the set (see E3) indicates that the input data (see E2) is the operational output data based on GCN deep neural network mechanics (see a) and the input data of the test learning is the test data (see B3).
With reference to fig. 1, 2 and 3, in the process of label pasting portrait classification output, a mobile phone WiFi probe 6 minimum matrix unit collection step based on label pasting portrait real-time work data, wherein the mobile phone WiFi probe 6 minimum matrix unit collection refers to the content in the foregoing steps, and includes the mobile phone WiFi probe 6 minimum matrix unit collection step based on real-time work data, the real-time work data is unmarked and unmarked small-town normal site mobile phone activity data, the mobile phone activity data is a mobile phone WiFi network card Mac address and intensity information data thereof collected by a mobile phone WiFi probe 6 minimum matrix unit 13 during small-town normal work, and the mobile phone is a mobile phone held by small-town store employees (i.e. waiters) 4, anonymous customers 10, anonymous passerbies 11 and anonymous snoopers 12;
in the present embodiment, the labeling portrait topological convolution operation-based real-time work data input and operation step refers to the content of the above steps, and includes a topological convolution operation-based real-time work data input and operation step, which is a process of inputting real-time work data via a conversion interface (see D2) and outputting an operation (see E2) via GCN deep neural network machine learning (see a) based on the above training, parameter adjustment, and test learning steps.
In the present embodiment, the step of outputting (see E5) the classification data calculated by the linear classification neural network (see E4) based on the labeling portrait, the classification data output (see E5) referring to the content of the preceding steps, includes the step of outputting (see E5) the classification data calculated by the linear classification neural network (see E4), the classification data output (see E5) calculated by the linear classification neural network (see E4), i.e., the input data (see E2), via the linear classification neural network (see E4), as the classification data output (see E5), the classification data output (see E5) is the output value number 1, 2, 3, the output value number 1, 2, 3 respectively represents the classification of the unlabeled real-time working data, i.e., the anonymous and unlabeled anonymous customer 10, the unlabeled real-time working data classification, The label prediction classification of the anonymous passerby 11 and the anonymous visitor 12, and the behavior prediction images of a visiting shop (see edge I), a visiting cabinet (see edge II) and a purchasing (see edge III).
With reference to fig. 1, 2 and 3, the cognitive matrix internet of things enables labeling portrait technology based on behavior prediction to realize behavior prediction data labeling, behavior prediction learning training and behavior prediction labeling portrait classification output; therefore, the invention realizes the personnel labeling and portrait process of the small town (system) from learning and understanding the interactive relation change knowledge based on the behavior prediction of the customers 10, the passerby 11, the visitor 12 and the waiter 4 in the area to the personnel labeling and portrait process based on the behavior prediction cognition of the shops (see edge I), the cabinets (see edge II) and the purchases (see edge III).
The construction method disclosed by the invention is combined with the construction methods shown in fig. 1, fig. 2, fig. 3 and fig. 4 to construct the internet of things of the cognitive matrix in the area (such as a town), and further comprises the construction of network cooperation technology energization based on space-time metering.
According to one embodiment of the invention, the step of enabling the spatio-temporal metering-based network cooperation technology comprises spatio-temporal metering analysis visualization and network cooperation scheduling of node services.
In the present embodiment, in the process of performing the visualization of the spatiotemporal metrology analysis, the time-space coordinate system setting step based on the town area is to convert the three-dimensional space area position coordinate system of the town into a two-dimensional plane area position coordinate system, where the two-dimensional plane area position coordinate system is a dimensional value of the areas in different zones in the same X-axis (see G1) and Y-axis (see G3) two-dimensional plane coordinate systems representing different floor areas as different magnitude.
In the embodiment, the panel data heat degree calculation presentation step based on the regional space-time measurement is to arrange the probe grid units of different places on different coordinate positions in the two-dimensional plane coordinate system and carry out measurement statistics on frequency values of events occurring to different people acquired by the probe according to time, and the panel data heat degree calculation presentation is to represent the frequency values of the calculation statistics obtained on sections (see G4) of different time coordinates (see T-axis values on G2) as heat values and present the frequency values as heat colors and express the heat colors on the sections (see G4) of the time coordinates.
In the present embodiment, the step of visualizing time dimension data analysis based on spatiotemporal measurements of regions, the time dimension data being data (see G5) plotted by calculating based on acquisition of coordinates x (see G6), y (see G8) at set positions (see G7) of a probe grid (see G2), the data analysis being based on data trend analysis of changes over time (see G2) of the plotted curve (see G5), the analysis visualization being a visualization of data trend analysis of changes over time (see G2) of the curve (see G5).
In the embodiment, in the process of performing network collaborative scheduling of node services, a node network collaborative step based on regional space-time measurement is performed, the regional space-time measurement is a panel (see G4) data heat analysis measurement, the node network is a node network formed by nodes of probe grid position diagram (see G7) of a regional space-time coordinate system (see G1, G2 and G3), the node network collaboration is a network scheduling service for balancing node overheating, and the balancing node overheating is a network scheduling operation for guiding the number of the balancing node excess to other nodes for balanced shunting operation.
According to the method, the time-space metering analysis visualization and the network cooperative scheduling of the node service are realized through the energization of the time-space metering-based network cooperative technology. Therefore, the invention realizes the node network cooperative service capability based on the space-time metering analysis through the technology.
The foregoing is merely exemplary of particular aspects of the present invention and devices and structures not specifically described herein are understood to be those of ordinary skill in the art and are intended to be implemented in such conventional ways.
The above is only one embodiment of the present invention, and is not intended to limit the present invention, and various modifications and changes may be made to the present invention by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (15)
1. A method for constructing a town cognition matrix Internet of things comprises the following steps:
s1, carrying out location association construction on the composition elements in the region based on a knowledge graph; wherein the constituent elements include: a business activity zone, a server post zone and a client activity zone; in step S1, associating the distribution space of the business activity location with the store business number based on the knowledge-graph; associating an attendant post location space in the business activity location with attendant information based on the knowledge-graph; carrying out location association on the server post location space and the client activity location space based on the knowledge map;
s2, constructing a probe grid for detecting the mobile equipment in the area;
s3, performing matrix acquisition construction for acquiring the mobile data of the mobile equipment in the area, and performing data labeling construction for labeling the mobile data;
and S4, carrying out topology convolution construction for machine training and learning and carrying out prediction classification construction based on the mobile data.
2. The building method according to claim 1, wherein step S2 includes:
s21, constructing a minimum triangular grid unit comprising three mobile equipment probes based on a topological structure;
s22, performing effective iteration based on the minimum triangular grid unit, and optimizing the minimum triangular grid unit according to an effective iteration result;
and S23, installing the mobile equipment probe based on the optimized vertex of the minimum triangular grid unit.
3. The building method according to claim 2, wherein in step S3, the step of performing matrix acquisition building of the movement data of the mobile device in the area comprises:
s31, detecting the mobile equipment based on a single mobile equipment probe in the minimum triangular grid unit and associating the acquired mobile data;
s32, detecting the mobile equipment based on the three mobile equipment probes in the minimum triangular grid unit and calculating the associated mobile data, wherein in the process of calculating the associated mobile data, the three mobile equipment probes perform associated positioning calculation on three different signal intensities of the same mobile equipment;
s33, establishing position track data records of all roles in the area, wherein the position track data records are position set data records of the role association time, and the roles comprise waiters, customers, passerby and visitors.
4. The construction method according to claim 3, wherein the step of constructing the data label labeling the mobile data in step S3 includes:
s34, performing attendant data annotation based on the attendant position track, wherein data records of the attendant position track are integrated into a data matrix or a data set associated with the attendant time position track;
s35, marking customer data based on the customer position track, wherein data records of the customer position track are integrated into a data matrix or a data set associated with the customer time position track;
s36, carrying out passenger data annotation based on the passenger position track, wherein data records of the passenger position track are integrated into a data matrix or a data set associated with the passenger time position track;
s37, carrying out passenger detection data annotation based on the passenger detection position track, wherein data records of the passenger detection position track are integrated into a data matrix or a data set associated with the passenger detection time position track.
5. The construction method according to claim 4, wherein the step of performing topology convolution construction for machine training learning based on the movement data in step S4 includes:
s41, performing data planning of machine learning training based on topological convolution, wherein the data planning is to convert data labels into machine learning training data in proportion through a label planning conversion interface, and the data labels are customer data labels, road passenger data labels, visitor data labels and waiter data labels respectively;
s42, training and learning of training data are carried out based on topological convolution, wherein the training and learning is to transfer the training data to GCN deep neural network machine learning based on a training input conversion interface, the GCN deep neural network machine learning is to output input topological matrix data to input data of a linear neural network through series of GCN topological convolution operations and a pooling conversion interface, and the series of GCN topological convolution operations are processes of outputting corresponding convolution operation data through convolution template operations;
s43, parameter adjusting learning of parameter adjusting data is carried out based on topological convolution, wherein the parameter adjusting learning is parameter optimization adjusting learning of the convolution template, the parameter optimization adjusting learning is convolution template parameter optimization adjusting learning based on back propagation gradient calculation, and input data of the convolution template parameter optimization adjusting learning is parameter adjusting data;
s44, test learning of test data is carried out based on topological convolution, wherein the test learning is the test learning of the output effect of the classification learning of the linear neural network added on the basis of the training learning and parameter adjusting learning steps, the classified data output effect test learning is to calculate tensor as input through a linear classification neural network into classified data output, the output value numbers of the classified data output are 1, 2 and 3, and the output value numbers 1, 2 and 3 represent the prediction classification of customers, passers-by and visitors of the test data respectively, the tensor is a tensor representation of the two-dimensional matrix eigen input data that is internally converted into a one-dimensional matrix eigen data set, the two-dimensional matrix characteristic input data is operation output data based on GCN deep neural network machine learning, and the input data of test learning is test data.
6. The construction method according to claim 5, wherein in step S41, the data planning is performed by scaling the data labels into the machine learning training data through the label planning conversion interface, wherein the data labels are scaled into the machine learning training data through the label planning conversion interface according to the proportion of 70% training data, 15% parameter adjustment data and 15% test data.
7. The constructing method according to claim 6, wherein the step of constructing the prediction classification in step S4 includes:
s45, acquiring activity data of all the mobile equipment in a minimum triangular grid unit based on real-time working data, wherein the real-time working data is the activity data of the mobile equipment in an unmarked area, and the activity data comprises the address data of a wireless network card of the mobile equipment and the signal intensity information data of the mobile equipment;
s46, performing real-time work data input and operation based on topological convolution operation, wherein the real-time work data input and operation is a process of inputting the real-time work data through a practice conversion interface and outputting the real-time work data through GCN deep neural network machine learning operation on the basis of the steps S42 to S44;
s47, classified data output is carried out based on linear classification neural network operation, namely input data are subjected to linear classification neural network operation and classified data output, the classified data output is that output value numbers are 1, 2 and 3, the output value numbers 1, 2 and 3 represent unmarked real-time working data classification, and the unmarked real-time working data classification is anonymous and unmarked anonymous customer, anonymous road passenger and anonymous passenger detection prediction classification.
8. The construction method according to claim 1 or 7, further comprising:
s5, enabling labeling portrait technology based on behavior prediction;
and S6, enabling a network cooperation technology based on space-time metering.
9. The building method according to claim 8, wherein step S5 includes:
s51, behavior prediction data marking, wherein the behavior prediction data marking comprises shop detection behavior prediction data marking based on the conversion of a passerby into a visitor, cabinet detection behavior prediction data marking based on the conversion of the visitor into a customer, and purchasing behavior prediction data marking based on the conversion of an old customer into new shopping;
s52, behavior prediction learning training, wherein the training data planning comprises machine learning training data planning based on behavior prediction topological convolution, training learning based on behavior prediction topological convolution training data, parameter adjusting learning based on behavior prediction topological convolution parameter adjusting data, and test learning based on behavior prediction topological convolution test data;
and S53, labeling portrait classification output, wherein the labeling portrait classification output comprises the steps of collecting the mobile equipment in a minimum triangular grid unit based on labeling portrait real-time working data, inputting and calculating the labeling portrait topology convolution operation based on real-time working data, and outputting classification data based on labeling portrait linear classification neural network operation.
10. The construction method according to claim 9, wherein in step S51, in the step of labeling the forecast data based on the probe store behavior converted from the road passenger, the label of the forecast data of the probe store behavior is a label of forecast data of the probe store relationship affinity and sparseness locus description converted from the server and the road passenger, the description of the probe store relationship affinity and sparseness locus is a calculation of a probe store difference value associated between the server position locus and the road passenger position locus, and the calculation of the probe store difference value is a calculation of grasping a characteristic of probe store association trend based on a matrix;
in the step of converting the visitor into the forecast data of the customer, the forecast data of the visitor is labeled, namely the forecast data of the visitor converted from the description of the visitor's visit cabinet relationship affinity and sparsity locus of the attendant and the visitor is labeled, the description of the visit cabinet relationship affinity and sparsity locus is the calculation of the visit cabinet difference value associated between the attendant's position locus and the visitor's position locus, and the calculation of the visit cabinet difference value locus is the calculation of the grabbing of the visit cabinet association trend characteristic based on the matrix;
in the step of converting old customers into purchasing behavior prediction data labels of new shopping, the purchasing behavior prediction data labels are obtained by converting purchasing relationship affinity and sparsity track descriptions of old customers and waiters into converted purchasing behavior prediction data labels, the purchasing relationship affinity and sparsity track descriptions are purchasing difference value calculation of the waiter position track and the old customer position track in association, and the purchasing difference value track calculation is purchasing association trend characteristic capture calculation based on a matrix.
11. The construction method according to claim 9, wherein in step S52, in the step of planning the machine learning training data based on the behavior prediction topological convolution, the planning of the machine learning training data is to transfer data labels into the machine learning training data in proportion through a label planning conversion interface, and the data labels are respectively a store exploration behavior prediction data label, a cabinet exploration behavior prediction data label, a purchase behavior prediction data label, and a server data label;
in the step of training and learning based on behavior prediction topological convolution training data, the training and learning is to transfer the training data to GCN deep neural network machine learning based on a training input conversion interface, the GCN deep neural network machine learning is to output input topological matrix data to input data of a linear neural network through series of GCN topological convolution operations and a pooling conversion interface, and the series of GCN topological convolution operations are processes of outputting corresponding convolution operation data through convolution template operations;
a parameter adjusting learning step of predicting topological convolution parameter adjusting data based on behaviors, wherein the parameter adjusting learning is parameter optimization adjustment learning of the convolution template, the parameter optimization adjustment learning is convolution template parameter optimization adjustment learning based on back propagation gradient calculation, and input data of the convolution template parameter optimization adjustment learning is parameter adjusting data;
a test learning step based on the behavior prediction topological convolution test data, wherein the test learning is a classification data output effect test learning which adds linear neural network classification learning on the basis of the training learning and the parameter adjusting learning, the classified data output effect test learning is to calculate tensor as input through a linear classification neural network into classified data output, the output value numbers of the classified data output are 1, 2 and 3, and the output value numbers are 1, 2 and 3, which represent the classification of the store visit behavior prediction, the cabinet visit behavior prediction and the purchase behavior prediction of the test data respectively, the tensor is a tensor representation of the two-dimensional matrix eigen input data that is internally converted into a one-dimensional matrix eigen data set, the two-dimensional matrix characteristic input data is operation output data based on GCN deep neural network machine learning, and the input data of test learning is test data.
12. The constructing method according to claim 9, wherein in the step of collecting the mobile device in the minimum triangular mesh unit based on the labeling portrait real-time working data in step S53, the labeling portrait real-time working data is the activity data of the mobile device in the unmarked and unmarked area, the activity data includes the wireless network card address data of the mobile device and the signal strength information data of the mobile device;
in the step of inputting and calculating the real-time working data based on the labeling portrait topological convolution operation, the real-time working data is input through the practice conversion interface and output through the GCN deep neural network machine learning operation on the basis of the step S52;
the step of outputting classification data based on labeling portrait linear classification neural network operation is a process that input data is subjected to linear classification neural network operation and classification data output, wherein the classification data output is that output value numbers are 1, 2 and 3, the output value numbers 1, 2 and 3 are unlabeled real-time working data classifications, and the unlabeled real-time working data classifications are anonymous and unlabeled labeling anonymous customer, anonymous passerby and anonymous visitor labeling prediction classifications, and a shop detection behavior prediction portrait, a cabinet detection behavior prediction portrait and a purchase behavior prediction portrait.
13. The building method according to claim 8, wherein step S6 includes:
s61, performing space-time metering analysis visualization, wherein the space-time metering analysis visualization comprises time-space coordinate system design, panel data heat degree calculation presentation and time dimension data analysis visualization;
and S62, network cooperative scheduling of node service.
14. The building method according to claim 13, wherein in step S61, the step of designing a time-space coordinate system is based on the time-space coordinate system design in the area, wherein the time-space coordinate system design is to convert a three-dimensional space area position coordinate system in the area into a two-dimensional plane area position coordinate system, and the two-dimensional plane area position coordinate system is to represent each floor area as a dimensional value of each area in the same two-dimensional plane coordinate system with different magnitude of values;
in the step of calculating and presenting panel data heat, calculating and presenting the panel data heat based on regional space-time measurement, wherein the regional space-time measurement is to arrange probe grid units of each place on each coordinate position in a two-dimensional plane coordinate system and measure and count frequency values of events of different people acquired by a probe according to time, and the panel data heat calculation presentation is to express the frequency values of calculation and statistics obtained on each time coordinate section as heat values, as heat colors and on the time coordinate section;
and time dimension data analysis visualization, wherein the time dimension data is data of a plotted curve acquired and calculated according to set time on the basis of a probe grid at set position coordinates, the time dimension data analysis is data change trend analysis of the plotted curve changing with time, and the data change trend analysis of the curve changing with time is visualized.
15. The construction method according to claim 13, wherein in step S62, in the step of network cooperation scheduling of node services, node network cooperation is performed based on area spatio-temporal measurements, the area spatio-temporal measurements are panel data heat analysis measurements, the node network is a node network composed of probe grid position nodes of an area spatio-temporal coordinate system, the node network cooperation is a network scheduling service for balancing node overheating, and the balancing node overheating is a network scheduling service for guiding the number of the balancing node excess to other nodes for balanced offloading operation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910022071.4A CN109729171B (en) | 2019-01-10 | 2019-01-10 | Method for constructing town cognitive matrix Internet of things |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910022071.4A CN109729171B (en) | 2019-01-10 | 2019-01-10 | Method for constructing town cognitive matrix Internet of things |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109729171A CN109729171A (en) | 2019-05-07 |
| CN109729171B true CN109729171B (en) | 2021-07-30 |
Family
ID=66299654
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910022071.4A Active CN109729171B (en) | 2019-01-10 | 2019-01-10 | Method for constructing town cognitive matrix Internet of things |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109729171B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111949840B (en) * | 2019-05-15 | 2024-05-31 | 杭州海康威视数字技术股份有限公司 | Method and device for constructing topological graph structure based on data of Internet of things |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7003502B1 (en) * | 2001-07-17 | 2006-02-21 | Unisys Corporation | Method for knowledge management |
| CN105824802A (en) * | 2016-03-31 | 2016-08-03 | 清华大学 | Method and device for acquiring knowledge graph vectoring expression |
| CN106815293A (en) * | 2016-12-08 | 2017-06-09 | 中国电子科技集团公司第三十二研究所 | System and method for constructing knowledge graph for information analysis |
| CN106919689A (en) * | 2017-03-03 | 2017-07-04 | 中国科学技术信息研究所 | Professional domain knowledge mapping dynamic fixing method based on definitions blocks of knowledge |
| CN108306962A (en) * | 2018-01-30 | 2018-07-20 | 河海大学常州校区 | A kind of business big data analysis system |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10219163B2 (en) * | 2013-03-15 | 2019-02-26 | DGS Global Systems, Inc. | Systems, methods, and devices for electronic spectrum management |
-
2019
- 2019-01-10 CN CN201910022071.4A patent/CN109729171B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7003502B1 (en) * | 2001-07-17 | 2006-02-21 | Unisys Corporation | Method for knowledge management |
| CN105824802A (en) * | 2016-03-31 | 2016-08-03 | 清华大学 | Method and device for acquiring knowledge graph vectoring expression |
| CN106815293A (en) * | 2016-12-08 | 2017-06-09 | 中国电子科技集团公司第三十二研究所 | System and method for constructing knowledge graph for information analysis |
| CN106919689A (en) * | 2017-03-03 | 2017-07-04 | 中国科学技术信息研究所 | Professional domain knowledge mapping dynamic fixing method based on definitions blocks of knowledge |
| CN108306962A (en) * | 2018-01-30 | 2018-07-20 | 河海大学常州校区 | A kind of business big data analysis system |
Non-Patent Citations (1)
| Title |
|---|
| "国内分子生物学知识图谱的构建及解读";乐露露;《CNKI》;20131215;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109729171A (en) | 2019-05-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Jelokhani-Niaraki et al. | A group multicriteria spatial decision support system for parking site selection problem: A case study | |
| Geertman et al. | Introduction to ‘planning support systems and smart cities’ | |
| Fang et al. | Spatiotemporal model for assessing the stability of urban human convergence and divergence patterns | |
| CN117892887A (en) | Land use optimizing system based on big data | |
| Brauer et al. | The state of the art in smart city research–A literature analysis on green IS solutions to foster environmental sustainability | |
| Stewart et al. | CoAXs: A collaborative accessibility-based stakeholder engagement system for communicating transport impacts | |
| CN117649027B (en) | Data processing method and system based on intelligent station | |
| CN112949997A (en) | System and method for community portrayal in urban planning design | |
| Strobel et al. | OMOD: An open-source tool for creating disaggregated mobility demand based on OpenStreetMap | |
| CN105893455A (en) | Method for forecasting conditions of locations and computer system | |
| CN109729171B (en) | Method for constructing town cognitive matrix Internet of things | |
| Zhu et al. | Prediction of consumer repurchase behavior based on LSTM neural network model | |
| Danchev et al. | Migration networks: applications of network analysis to macroscale migration patterns | |
| Li et al. | Uncovering the relationship between point-of-interests-related human mobility and socioeconomic status | |
| Luo et al. | TransFlower: An explainable transformer-based model with flow-to-flow attention for commuting flow prediction | |
| CN107436945A (en) | A kind of information collecting method and system based on electronic map | |
| Kondor et al. | Visualizing signatures of human activity in cities across the globe | |
| Rathod et al. | Composite accessibility index: a novel and holistic measure for evaluating transit accessibility | |
| Schoemakers et al. | LUTI modelling in the Netherlands: Experiences with TIGRIS and a framework for a new LUTI model | |
| Amalan et al. | High frequency data in Land use and transport integrated model: A review of sources and application | |
| Cordeiro et al. | Proposal of Maturity Assessment to Support Projects in Digital Transformation: Initiatives in the Brazilian Context | |
| Savchenko-Synyakova et al. | The Tools for Intelligent Data Analysis, Modeling and Forecasting Of Social and Economic Processes | |
| Stonor | Space Syntax: a SMART approach to urban planning, design and governance | |
| KR102832271B1 (en) | Methods of providing event planning and operation services | |
| Zhang | Accounting computerization data adjustment method on account of artificial intelligence algorithm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20231214 Address after: Room 1702, Building 1, Haiyun International Building, Shangcheng District, Hangzhou City, Zhejiang Province, 310000 Patentee after: Qicai Engineering Consulting (Hangzhou) Co.,Ltd. Address before: Room 284-2, 857 shixinbei Road, ningwei street, Xiaoshan District, Hangzhou City, Zhejiang Province Patentee before: EASI WISDOM TECHNOLOGY Co.,Ltd. |