CN110019401A - Part amount prediction technique, device, equipment and its storage medium - Google Patents
Part amount prediction technique, device, equipment and its storage medium Download PDFInfo
- Publication number
- CN110019401A CN110019401A CN201711426164.0A CN201711426164A CN110019401A CN 110019401 A CN110019401 A CN 110019401A CN 201711426164 A CN201711426164 A CN 201711426164A CN 110019401 A CN110019401 A CN 110019401A
- Authority
- CN
- China
- Prior art keywords
- data
- model
- time
- prediction
- obtains
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 40
- 238000003860 storage Methods 0.000 title claims abstract description 19
- 238000012549 training Methods 0.000 claims abstract description 95
- 238000012545 processing Methods 0.000 claims abstract description 92
- 238000006116 polymerization reaction Methods 0.000 claims description 52
- 238000010200 validation analysis Methods 0.000 claims description 27
- 238000013136 deep learning model Methods 0.000 claims description 26
- 238000013480 data collection Methods 0.000 claims description 15
- 238000012795 verification Methods 0.000 claims description 14
- 230000000737 periodic effect Effects 0.000 claims description 10
- 238000004590 computer program Methods 0.000 claims description 7
- 235000013399 edible fruits Nutrition 0.000 claims 2
- 238000005096 rolling process Methods 0.000 abstract description 9
- 238000012423 maintenance Methods 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 14
- 238000013135 deep learning Methods 0.000 description 8
- 238000000605 extraction Methods 0.000 description 7
- 238000011161 development Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 6
- 230000007774 longterm Effects 0.000 description 6
- PEDCQBHIVMGVHV-UHFFFAOYSA-N Glycerine Chemical compound OCC(O)CO PEDCQBHIVMGVHV-UHFFFAOYSA-N 0.000 description 5
- 238000002790 cross-validation Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 238000009825 accumulation Methods 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 4
- 230000006854 communication Effects 0.000 description 4
- 238000013507 mapping Methods 0.000 description 4
- 238000003062 neural network model Methods 0.000 description 4
- 238000002360 preparation method Methods 0.000 description 4
- 230000006403 short-term memory Effects 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 241001269238 Data Species 0.000 description 3
- 238000004891 communication Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 235000007926 Craterellus fallax Nutrition 0.000 description 2
- 240000007175 Datura inoxia Species 0.000 description 2
- 150000001875 compounds Chemical class 0.000 description 2
- 238000007418 data mining Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000015654 memory Effects 0.000 description 2
- 230000001737 promoting effect Effects 0.000 description 2
- 230000035484 reaction time Effects 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000013277 forecasting method Methods 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
- G06F16/2465—Query processing support for facilitating data mining operations in structured databases
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
- G06F16/2474—Sequence data queries, e.g. querying versioned data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
- G06F16/2477—Temporal data queries
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2216/00—Indexing scheme relating to additional aspects of information retrieval not explicitly covered by G06F16/00 and subgroups
- G06F2216/03—Data mining
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Databases & Information Systems (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Fuzzy Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
This application discloses part amount prediction technique, device, equipment and its storage mediums.This method comprises: within the scope of first time history waybill data and external data handle, obtain data processed result；It is trained using data processed result, obtains depth residual error network foundation model；The basic model is updated using the history waybill data in the second time range, obtains prediction model to determine the prediction part amount of target wave time.Technical solution provided by the embodiments of the present application overcomes and individually models the problem for causing maintenance cost excessively high for single site in the prior art；And by the dimension-reduction treatment to history waybill data, after period, trend study processing, training obtains part amount of the depth residual error network model to predict non-incoming wave time, improves the precision of prediction result.Basic model also is updated by way of rolling more new data, to greatly reduce consumption and the time overhead of computing resource.
Description
Technical field
Present application relates generally to computer fields, and in particular to data mining processing technology field more particularly to part amount are pre-
Survey method, apparatus, equipment and its storage medium.
Background technique
Important tie of the logistics field as connection socio-economic development and social life, the data mining based on the field
Belong to emerging research field, the development of big data technology brings new opportunity to logistic industry, reasonably uses big data
Technology plays a positive role to management and the decision of logistic industry, customer relationship maintenance, resource distribution etc..
In the prior art, the recurrent neural network in time series models (recurrent neural network, RNN)
It applies in logistic industry, sends part amount information for predict site.For example, the time sequence forecasting method based on Generalized Additive Models can
It is predicted with the part amount daily to each site, but this prediction mode has that data volume is big there are wave time part amount prediction,
Part amount mean value is small, the big problem of prediction error.And that there are dot datas is more for existing prediction mode, separately modeling will lead to number
According to the problems such as sparse, maintenance cost is excessively high.Furthermore, it is contemplated that the factors such as festivals or holidays, extreme weather and time dynamic
It influences, the model established based on dot data, it is understood that there may be data dithering is larger, and model performance is unstable, and models speed
The problems such as degree is slow time-consuming high.
It would therefore be highly desirable to propose the new prediction model of one kind to solve the above problems.
Summary of the invention
In view of drawbacks described above in the prior art or deficiency, it is intended to provide a kind of based on depth residual error neural network forecast target wave
The scheme of secondary part amount.
In a first aspect, the embodiment of the present application provides a kind of amount prediction technique, this method comprises:
To within the scope of first time history waybill data and external data handle, obtain data processed result；
It is trained using data processed result, obtains depth residual error network foundation model；
The basic model is updated using the history waybill data in the second time range, obtains prediction model to determine target
The prediction part amount of wave time.
Second aspect, the embodiment of the present application provide a kind of amount prediction meanss, which includes:
Data processing unit, within the scope of first time history waybill data and external data handle, obtain
To data processed result；
It creates basic model unit and obtains depth residual error network foundation mould for being trained using data processed result
Type；
Predicting unit is updated to obtain for updating the basic model using the history waybill data in the second time range
Prediction model determines the prediction part amount of target wave time.
The third aspect, the embodiment of the present application provide a kind of equipment, including processor, storage device；
Aforementioned storage device, for storing one or more programs；
When aforementioned one or more programs are executed by aforementioned processor, so that aforementioned processor realizes that the embodiment of the present application is retouched
The method stated.
Fourth aspect, the embodiment of the present application provide a kind of computer readable storage medium, are stored thereon with computer journey
Sequence when aforementioned computer program is executed by processor, realizes the method that the embodiment of the present application describes.
Part amount prediction scheme provided by the embodiments of the present application, by relevant to wave time part amount prediction to data processing extraction
Important attribute data, and learn wave secondary period, trend characteristic from history waybill data using LSTM model, and for wave
The secondary influential external data of part amount, the polymerization training depth residual error network foundation model based on these information, which can
The training time is reduced, the efficiency of lift scheme creation.Further, basic model is updated by way of rolling forecast also to predict
The part amount of target wave time can relatively accurately predict following wave time part amount.The embodiment of the present application, it is also multiple not by training
Same more new model carries out cross validation, the stability of Lai Tisheng prediction result.
According to the technical solution of the embodiment of the present application, overcoming individually to model for single site in the prior art causes to safeguard
The excessively high problem of cost, and by the dimension-reduction treatment to history waybill data, after period, trend study processing, training obtains depth
Part amount of the degree residual error network model to predict non-incoming wave time, improves the precision of prediction result.Also number is updated by rolling
According to mode update basic model, greatly reduce consumption and the time overhead of computing resource.
Detailed description of the invention
By reading a detailed description of non-restrictive embodiments in the light of the attached drawings below, the application's is other
Feature, objects and advantages will become more apparent upon:
Fig. 1 shows the flow diagram of part amount prediction technique provided by the embodiments of the present application；
Fig. 2 shows the flow diagrams for the part amount prediction technique that the another embodiment of the application provides；
Fig. 3 shows the schematic illustration of Data Dimensionality Reduction processing；
Fig. 4 shows the schematic illustration of periodic law and recent trend study；
Fig. 5 shows the schematic illustration of single time series study；
Fig. 6 shows the schematic illustration of depth residual error network foundation model；
Fig. 7 shows the structural schematic diagram of part amount prediction meanss provided by the embodiments of the present application；
Fig. 8 shows the structural schematic diagram for the part amount prediction meanss that the another embodiment of the application provides；
Fig. 9 shows the structural schematic diagram for being suitable for the computer system for the terminal device for being used to realize the embodiment of the present application.
Specific embodiment
The application is described in further detail with reference to the accompanying drawings and examples.It is understood that this place is retouched
The specific embodiment stated is used only for explaining related invention, rather than the restriction to the invention.It also should be noted that in order to
Convenient for description, part relevant to invention is illustrated only in attached drawing.
It should be noted that in the absence of conflict, the features in the embodiments and the embodiments of the present application can phase
Mutually combination.The application is described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.
Referring to FIG. 1, Fig. 1 shows the flow diagram of part amount prediction technique provided by the embodiments of the present application.
As shown in Figure 1, this method comprises:
Step 101, within the scope of first time history waybill data and external data handle, obtain data processing
As a result.
In the embodiment of the present application, in data preparation stage, the history waybill data and external number of setting time range are obtained
According to, and be used to establish model after handling these data.The data resource of logistic industry, can be from logistics number compared with horn of plenty
It is handled according to middle extraction history waybill data.In order to overcome dot data amount big, part amount mean value is small, and prediction error is big to ask
Topic can create model with all dot datas of a city (or business district), predict site by unified Modeling
Wave time part amount.
Setting time may range from first time range, and the value of first time range can be to be set as unit of year
It sets.For example, with 2 years history waybill data instances.2 years history waybill data are obtained, extracting influences in history waybill data
The characteristic attribute data of wave time part amount, and it is used to training pattern after handling characteristic attribute data.Wherein, characteristic attribute number
According to including: number node data, month data, day data, week data, wave time part amount data etc..
Dimension-reduction treatment is carried out to characteristic attribute data, helps to extract representative strong characteristic attribute.By dimension-reduction treatment
Data afterwards are more advantageous to extraction effective information, improve the efficiency of data processing convenient for calculating and visualizing.
The linear mapping of the normal method of dimension-reduction treatment and Nonlinear Mapping.For example, can be used in deep learning model
Embedding mode processing feature attribute data, is mapped to feature space.
The embodiment of the present application learns history waybill data using deep learning model, so that obtaining can embody
The time series of data development trend.The part amount data of site have stronger periodicity, are transported using deep learning model from history
This regularity is extracted in forms data, the wave time part amount of some following period is predicted for creating basic model.
For example, learning using deep learning model to history waybill data, the time sequence of rule reaction time is extracted
The time series of column and recent trend.Cycle time sequence, the time series such as unit of week, trend time series, such as 28
Its trend time series.Deep learning model in the embodiment of the present application can be deep learning shot and long term memory models
(LSTM) model or other neural network models and other can be used in the model of learning data rule.Preferably,
The embodiment of the present application selected depth Chief Learning Officer, CLO's short-term memory model (LSTM) model.
The embodiment of the present application, it is contemplated that part has also been introduced in modeling process to the influence degree of part amount in external factor
External data enhances the precision of model.For example, technical dates data, weather data etc..
Step 102, it is trained using data processed result, obtains depth residual error network foundation model.
The embodiment of the present application constructs model using result after data processing.It is obtaining after data processing
Result training pattern before, to data, treated as a result, it is desirable to carry out polymerization processing, and the result of polymerization processing is input to
It is trained in the deep learning model of building.For example, by Jing Guo dimension-reduction treatment characteristic attribute data and periodic law when
Between the time series and external data of sequence and recent trend carry out polymerization processing, obtain the result of polymerization processing.Utilize polymerization
The result of processing trains deep learning model.The embodiment of the present application, it is preferable that with depth residual error network model be basic mould
Type, the basic model may include multiple residual error modules.In the embodiment of the present application, pass through the less neural network of residual error network
Layer obtains preferably as a result, reducing the training time.
Step 103, aforementioned base model is updated using the history waybill data in the second time range, obtains prediction model
To determine the prediction part amount of target wave time.
The embodiment of the present application, using more new data come re -training basic model, obtains after creating basic model
New prediction model, to predict the wave time part amount of future time section using prediction model.Optionally, according to rolling forecast
Mode more new data.For example, the data group compound training data set that fetching portion history part amount data and part newly obtain, to base
Plinth model is updated month by month, to predict the wave time part amount data of some following period.
Wherein, the definition of wave time, it can be understood as the operation batch in the unit time, the unit time can be every 30 points
Zhong Weiyi wave is also possible to be divided according to the operation or work standard of practical industry.According to the different classes of of waybill, to wave time data
The mode handled is different.For example, waybill type, which can be divided into, sends part class, there are order addressee class, no order addressee class etc.,
In, wave defined in part class time to be sent, will be divided within whole day 24 hours 35 waves, the 1st wave time is 0 point to 7 points of morning, the 2nd
A wave time is 7 points to 7:30 minute, and the 3rd wave time assigns at 8 points for 7:30, and so on to generate a wave every half an hour secondary, one
It assigns at 0 point until the 35th wave is secondary for 23:30, the wave time data for sending part class is handled, so that part amount data more meet industry
Be engaged in scene, for example, by the part amount data accumulation of the previous day 21 points to 9 points of the next morning of each wave time to second day the 5th
A wave.
For another example, there is wave defined in order addressee class, it is similar to part class is sent, 35 waves were divided by whole day 24 hours,
But the wave for having order addressee class time data are handled, for example, by the 4th of the 31st wave of the previous day secondary to second day the
The part amount data accumulation of a wave time to second day the 4th wave is secondary.For another example, no order addressee class, it is similar to part class is sent, by whole day
35 waves are divided within 24 hours, still, the wave time data of no order addressee class are handled, for example, by the previous day the 35th
The 26th wave time that a wave time is evenly distributed to the previous day to the secondary part amount data of second day the 3rd wave is secondary to the 34th wave.
Prediction of the embodiment of the present application by deep learning to wave time part amount, there is performance preferably, can be more accurate
Ground predicts the part amount of each wave time, greatly improves prediction effect.The embodiment of the present application is remembered using deep learning shot and long term
Network (LSTM) model excavates the changing pattern in time series, both considers long term variations, while considering that Recent Changes become
Gesture reduces the model training time, improves the efficiency of model creation by creating depth residual error network foundation model.And lead to
The mode for crossing rolling forecast updates basic model, and it is slow, time-consuming to overcome conventional time series model creation speed in the prior art
High problem.
Referring to FIG. 2, the flow diagram of the part amount prediction technique provided Fig. 2 shows the another embodiment of the application.
As shown in Fig. 2, this method comprises:
Step 201, within the scope of first time history waybill data and external data handle, obtain data processing
As a result.
In the embodiment of the present application, in data preparation stage, the history waybill data and external number of setting time range are obtained
According to, and be used to establish model after handling these data.Setting time may range from first time range, at the first time
The value of range can be to be arranged as unit of year.
For history waybill data, the characteristic attribute data for influencing wave time part amount are obtained, are needed from history waybill data
Middle extraction characteristic attribute data, and characteristic attribute data are handled, obtain the first processing result.
For example, with 2 months -2017 years 2 months 2015 history waybill data instances.From the history waybill data of this period
It is middle to extract the characteristic attribute data for influencing wave time part amount.Wherein, characteristic attribute data include: number node data, month data,
Day data, week data, wave time part amount data etc..Optionally, as shown in figure 3, by respectively by number node data, month
Polymerization generates identification information after data, day data, week data, wave time part amount data carry out embedding processing respectively, i.e., in fact
The dimension-reduction treatment of existing characteristic attribute data.
For history waybill data, the data variation rule characteristic for influencing wave time part amount is obtained, is needed based on depth
It practises model and periodic law study is carried out to institute's history waybill data, obtain multiple first time sequences, and when by institute multiple first
Between sequential polymerization be cycle time sequence；Equally, the history waybill data are carried out based on aforementioned depth learning model recent
Trend study, obtains trend time series.Deep learning model in the embodiment of the present application can be deep learning shot and long term note
Recall model (LSTM) model or other neural network models and other can be used in the model of learning data rule.It is excellent
Selection of land, the embodiment of the present application selected depth Chief Learning Officer, CLO's short-term memory model (LSTM) model.
For example, excavating time series using LSTM model to 2 months -2017 years 2 months 2015 history waybill data with anti-
Reflect the cycle variation law and Near-future Development Trend rule of data.As shown in figure 4, history waybill data are input to LSTM model
In, it extracts history cycle time series and carries out polymerization as cycle time sequence.History waybill data are input to LSTM model,
Historical trending time sequence is extracted as trend time series.
Wherein, history cycle time series is extracted using history waybill data, as shown in figure 5, if extracting history waybill
The time series on some Monday temporally recalls mode on the basis of the Monday in data, before extracting the Monday
12 weeks Monday data are separately input to be handled in LSTM module, after 64 layers of full articulamentum processing, are output to
32 layers of full articulamentum processing, to generate the time series on the Monday.
Wherein, trend time series is extracted using history waybill data, as shown in figure 5, before specified time being extracted
28 days data generate trend time series.
The embodiment of the present application introduces external data in data processing stage further to analyze the variation of wave time part amount.It is right
External data is identified processing, obtains second processing result.
There are many modes for obtaining external data.For example, being extracted from internal data, or obtained by external data resource
It takes.It to the external data of acquisition, needs to be further processed, is converted into the feature for influencing wave time part amount.The application is implemented
The external data extracted in example can be technical dates data and weather data.Wherein, technical dates includes that the legal section in part is false
Day and part technical dates, for example, New Year's Day, the Spring Festival, the Dragon Boat Festival, Clear and Bright, May Day, mid-autumn, National Day, double 11, double ten second-class.To spy
The processing of different date data, it can be understood as with New Year's Day, the Spring Festival, the Dragon Boat Festival, Clear and Bright, May Day, mid-autumn, National Day, double 11, double 12
For column, date data is row, and the value range of data is [- 15,16] in each column.It is with the Spring Festival (i.e. the lunar New Year's Day of the lunar calendar)
Example, use 16 indicate 15 days before and after the date be not the Spring Festival, -15 indicate that there are also 15 days be the Spring Festival, and on the day of 0 indicates the Spring Festival, 15 are indicated
The Spring Festival passes by 15 days.By being identified processing to technical dates, technical dates and wave time part amount can be preferably excavated
Between influence relationship.
Processing for weather data, it can be understood as the case where indicating weather with numerical value, such as indicate the numerical value of weather
Range is { -2,0,2 }, wherein -2 indicate exceedingly odious weather, 2 indicate sunny, mends 0 when uncertain or normal.It is right
The processing of weather data, in order to preferably analyze influence of the weather conditions to wave time part amount.
After handling technical dates data and weather data, polymerization generates external data.
Optionally, step 201 can also include:
Step 2011, characteristic attribute data are extracted from history waybill data, and characteristic attribute data are handled, and are obtained
To the first processing result；
Step 2012, periodic law study is carried out to history waybill data based on deep learning model, obtains multiple first
Time series, and be cycle time sequence by multiple first time sequential polymerizations；
Step 2013, recent trend study is carried out to history waybill data based on deep learning model, obtains the trend time
Sequence；
Step 2014, processing is identified to external data, obtains second processing as a result, said external data include special
Date data and/or weather data.
Step 202, it is trained using data processed result, obtains depth residual error network foundation model.
Data processed result is obtained after handling history part amount data, data processed result polymerization is handled, is used
In training depth residual error network foundation model.
In the embodiment of the present application, by the first processing result, cycle time sequence, trend time series and second processing result
It is polymerize, obtains polymerization result.Polymerization result is input to depth residual error network model, is trained.The depth residual error net
Network model may include multiple residual error modules.As shown in fig. 6, by the first processing result, cycle time sequence, trend time series
Depth residual error network foundation model is input to the polymerization of second processing result.The depth residual error network foundation model includes 3 residual
Difference module.Each residual error module is built-up by a full articulamentum progress residual error study.
Optionally, step 202, comprising:
Step 2021, aggregated data processing result obtains polymerization result；
Step 2022, it is trained using polymerization result, obtains depth residual error network foundation model.
Wherein step 2021 further include: by processing result, multiple time serieses, the second time series and second processing result
Polymerization, obtains polymerization result.
Step 203, aforementioned base model is updated using the history waybill data in the second time range, obtains prediction model
To determine the prediction part amount of target wave time.
The time of the existing each every wave in site time of time series models training is longer, and generally higher than 20 minutes, daily
About 30 waves of wave time.As it can be seen that the training time that existing time series models expend is too long, lead to the forecasting efficiency of model
It is lower.For the variation for more effectively predicting non-incoming wave time part amount, need to construct the better model of more new capability, for promoting work
Efficiency.
The embodiment of the present application constructs depth residual error network foundation model, to the basic model, Ke Yitong in step 202
The mode of setting rolling forecast is crossed, to update the basic model.
For example, updating the basic model using the history waybill data in the second time range.Second time range is set
Setting can adjust according to data more new state dynamic.For example, prediction target is in March, 2017, then the second time range can be set
It is set to 2 months -2017 years in October, 2016.If prediction target is in April, 2017, the second time range can be set to 2016
In March, -2017 in November year.Alternatively, further reducing time range, such as predict that target is in March, 2017, then the second time model
It encloses and can be set to 2 months -2017 years in December, 2016.
The selection of the second time range is the mode monthly rolled in the embodiment of the present application, chooses the model of setting time span
It encloses.The time span can be the integral multiple of calendar month.For example, 3 months or 5 months.
After the history waybill data for obtaining the second time range, by the history waybill data further division of this period
At training dataset and validation data set.And re -training is carried out to basic model to update basic mould using training dataset
Type inputs updated basic model based on validation data set to obtain the wave of prediction time part amount, by dividing training dataset
With the mode of validation data set cross validation, to prevent updated model from overfitting problem occur.
Step 203, optionally, comprising:
Step 2031, the history waybill data of the second time range of prediction target wave time are determined；
Step 2032, the history waybill data of the second time range are divided into training dataset and validation data set；
Step 2033, using training dataset re -training depth residual error network foundation model, prediction model is obtained；
Step 2034, validation data set is input to prediction model, obtains the prediction part amount of target wave time.
For example, the second time range is 5 months by taking the wave time part amount for predicting in March, 2017 as an example, that is, choose 2016 10
The history waybill data of 2 months -2017 years moon.History waybill data are divided into training dataset and validation data set.
For example, { in January, -2017 in October, 2016 } is training dataset, { 2 months 2017 } are validation data set；
Basic model is trained with training dataset, the basic model updated, indicates referred to as pre- for difference
Survey model；Validation data set is input to prediction model, obtains prediction result, as the prediction part amount of target wave time.
Optionally, the embodiment of the present application by adjusting training set and is tested in order to further enhance the stability of prediction result
The multiple basic models of training for demonstrate,proving data set carry out cross validation to realize the prediction of wave time part amount.
Optionally, step 2032, can also include:
In the history waybill data of the second time range, multiple validation data sets are dynamically specified, and mark off multiple
With the one-to-one training dataset of the validation data set.
Optionally, step 2033, can also include:
Multiple training datasets are inputted into depth residual error network foundation model re -training respectively, obtain multiple and training number
Depth residual error Network Prediction Model is corresponded according to collection.
Optionally, step 2034, can also include:
It will be inputted respectively with the one-to-one validation data set of multiple training datasets a pair of with multiple training datasets one
The depth residual error network model answered obtains multiple prediction part amounts；
Calculate the average value of multiple prediction part amounts, the prediction part amount as target wave time.
For example, the second time range is 5 months by taking the wave time part amount for predicting in March, 2017 as an example, that is, choose 2016 10
The history waybill data of 2 months -2017 years moon.
The history waybill data in 2 months -2017 years in October, 2016 are divided into 3 kinds of different training datasets and verifying number
According to collection.
For example, { in January, -2017 in October, 2016 } is the first training dataset, { 2 months 2017 } are first verification data
Collection；
{ in October, -2016 in December, 2016,2 months 2017 } is the second training dataset, and { in January, 2017 } tests for second
Demonstrate,prove data set；
{ October in November, 2016-,-2 months in January, 2017 } is third training dataset, and { in December, 2016 } tests for third
Demonstrate,prove data set.
Basic model is trained with the first training dataset, the basic model updated, indicates to be referred to as difference
For the first prediction model；First verification data collection is input to the first prediction model, obtains the first prediction result.
Basic model is trained with the second training dataset, the basic model updated, indicates to be referred to as difference
For the second prediction model；First verification data collection is input to the second prediction model, obtains the second prediction result.
Basic model is trained with third training dataset, the basic model updated, indicates to be referred to as difference
For third prediction model；First verification data collection is input to the second prediction model, obtains third prediction result.
Finally, the first prediction result, the second prediction result and the second prediction result are averaged, final prediction is obtained
As a result.
It should be noted that although describing the operation of the method for the present invention in the accompanying drawings with particular order, this is not required that
Or hint must execute these operations in this particular order, or have to carry out operation shown in whole and be just able to achieve the phase
The result of prestige.On the contrary, the step of describing in flow chart can change and execute sequence.For example, based on deep learning model to history
Waybill data carry out periodic law study, obtain multiple first time sequences, and are the period by multiple first time sequential polymerizations
The step of time series, and recent trend study is carried out to history waybill data based on deep learning model, obtain the trend time
The step of sequence.Additionally or alternatively, it is convenient to omit multiple steps are merged into a step and executed by certain steps, and/or
One step is decomposed into execution of multiple steps.For example, being trained using data processed result, depth residual error network base is obtained
Plinth model, comprising: aggregated data processing result obtains polymerization result；It is trained using polymerization result, obtains depth residual error net
Network basic model.
Referring to FIG. 7, Fig. 7 shows the structural schematic diagram of part amount prediction meanss provided by the embodiments of the present application.
As shown in fig. 7, the device 700 includes:
Data processing unit 701, within the scope of first time history waybill data and external data handle,
Obtain data processed result.
In the embodiment of the present application, in data preparation stage, the history waybill data and external number of setting time range are obtained
According to, and be used to establish model after handling these data.The data resource of logistic industry, can be from logistics number compared with horn of plenty
It is handled according to middle extraction history waybill data.In order to overcome the problems, such as that dot data is sparse, can with a city (or business
Area) all dot datas create model, predict by unified Modeling the wave time part amount of site.
Setting time may range from first time range, and the value of first time range can be to be set as unit of year
It sets.For example, with 2 years history waybill data instances.2 years history waybill data are obtained, extracting influences in history waybill data
The characteristic attribute data of wave time part amount, and it is used to training pattern after handling characteristic attribute data.Wherein, characteristic attribute number
According to including: number node data, month data, day data, week data, wave time part amount data etc..
Dimension-reduction treatment is carried out to characteristic attribute data, helps to extract representative strong characteristic attribute.By dimension-reduction treatment
Data afterwards are more advantageous to extraction effective information, improve the efficiency of data processing convenient for calculating and visualizing.
The linear mapping of the normal method of dimension-reduction treatment and Nonlinear Mapping.For example, can be used in deep learning model
Embedding mode processing feature attribute data, is mapped to feature space.
The embodiment of the present application learns history waybill data using deep learning model, so that obtaining can embody
The time series of data development trend.The part amount data of site have stronger periodicity, are transported using deep learning model from history
This regularity is extracted in forms data, the wave time part amount of some following period is predicted for creating basic model.
For example, learning using deep learning model to history waybill data, the time sequence of rule reaction time is extracted
The time series of column and recent trend.The time series of periodic law, the time series such as unit of week, recent trend when
Between sequence, the time series such as unit of the moon.Deep learning model in the embodiment of the present application, can be deep learning length
Phase memory models (LSTM) model or other neural network models and other can be used in the mould of learning data rule
Type.Preferably, the embodiment of the present application selected depth Chief Learning Officer, CLO short-term memory model (LSTM) model.
The embodiment of the present application, it is contemplated that part has also been introduced in modeling process to the influence degree of part amount in external factor
External data enhances the precision of model.For example, technical dates data, weather data etc..
It creates basic model unit 702 and obtains depth residual error network foundation for being trained using data processed result
Model.
The embodiment of the present application constructs model using result after data processing.It is obtaining after data processing
Result training pattern before, to data, treated as a result, it is desirable to carry out polymerization processing, and the result of polymerization processing is input to
It is trained in the deep learning model of building.For example, by Jing Guo dimension-reduction treatment characteristic attribute data and periodic law when
Between the time series and external data of sequence and recent trend carry out polymerization processing, obtain the result of polymerization processing.Utilize polymerization
The result of processing trains deep learning model.The embodiment of the present application, it is preferable that with depth residual error network model be basic mould
Type, the basic model may include in multiple residual error module the embodiment of the present application by the less neural net layer of residual error network
It obtains preferably as a result, reducing the training time.
Predicting unit 703 is updated, for updating aforementioned base model using the history waybill data in the second time range,
Prediction model is obtained to determine the prediction part amount of target wave time.
The embodiment of the present application, using more new data come re -training basic model, obtains after creating basic model
New prediction model, to predict the wave time part amount of future time section using prediction model.Optionally, according to rolling forecast
Mode more new data.For example, the data group compound training data set that fetching portion history part amount data and part newly obtain, to base
Plinth model is updated month by month, to predict the wave time part amount data of some following period.
Wherein, the definition of wave time, it can be understood as the operation batch in the unit time, the unit time can be every 30 points
Zhong Weiyi wave is also possible to be divided according to the operation or work standard of practical industry.According to the different classes of of waybill, to wave time data
The mode handled is different.For example, waybill type, which can be divided into, sends part class, there are order addressee class, no order addressee class etc.,
In, wave defined in part class time to be sent, will be divided within whole day 24 hours 35 waves, the 1st wave time is 0 point to 7 points of morning, the 2nd
A wave time is 7 points to 7:30 minute, and the 3rd wave time assigns at 8 points for 7:30, and so on to generate a wave every half an hour secondary, one
It assigns at 0 point until the 35th wave is secondary for 23:30, the wave time data for sending part class is handled, so that part amount data more meet industry
Be engaged in scene, for example, by the part amount data accumulation of the previous day 21 points to 9 points of the next morning of each wave time to second day the 5th
A wave.
For another example, there is wave defined in order addressee class, it is similar to part class is sent, 35 waves were divided by whole day 24 hours,
But the wave for having order addressee class time data are handled, for example, by the 4th of the 31st wave of the previous day secondary to second day the
The part amount data accumulation of a wave time to second day the 4th wave is secondary.For another example, no order addressee class, it is similar to part class is sent, by whole day
35 waves are divided within 24 hours, still, the wave time data of no order addressee class are handled, for example, by the previous day the 35th
The 26th wave time that a wave time is evenly distributed to the previous day to the secondary part amount data of second day the 3rd wave is secondary to the 34th wave.
Prediction of the embodiment of the present application by deep learning to wave time part amount, there is performance preferably, can be more accurate
Ground predicts the part amount of each wave time, greatly improves prediction effect.The embodiment of the present application is remembered using deep learning shot and long term
Network (LSTM) model excavates the changing pattern in time series, both considers long term variations, while considering that Recent Changes become
Gesture reduces the model training time, improves the efficiency of model creation by creating depth residual error network foundation model.And lead to
The mode for crossing rolling forecast updates basic model, and it is slow, time-consuming to overcome conventional time series model creation speed in the prior art
High problem.
Referring to FIG. 8, Fig. 8 shows the structural schematic diagram for the part amount prediction meanss that the another embodiment of the application provides.
As shown in figure 8, the device 800 includes:
Data processing unit 801, within the scope of first time history waybill data and external data handle,
Obtain data processed result.
In the embodiment of the present application, in data preparation stage, the history waybill data and external number of setting time range are obtained
According to, and be used to establish model after handling these data.Setting time may range from first time range, at the first time
The value of range can be to be arranged as unit of year.
For history waybill data, the characteristic attribute data for influencing wave time part amount are obtained, are needed from history waybill data
Middle extraction characteristic attribute data, and characteristic attribute data are handled, obtain the first processing result.
For example, with 2 months -2017 years 2 months 2015 history waybill data instances.From the history waybill data of this period
It is middle to extract the characteristic attribute data for influencing wave time part amount.Wherein, characteristic attribute data include: number node data, month data,
Day data, week data, wave time part amount data etc..Optionally, as shown in figure 3, by respectively by number node data, month
Polymerization generates identification information after data, day data, week data, wave time part amount data carry out embedding processing respectively, i.e., in fact
The dimension-reduction treatment of existing characteristic attribute data.
For history waybill data, the data variation rule characteristic for influencing wave time part amount is obtained, is needed based on depth
It practises model and periodic law study is carried out to institute's history waybill data, obtain multiple first time sequences, and when by institute multiple first
Between sequential polymerization be cycle time sequence；Equally, the history waybill data are carried out based on aforementioned depth learning model recent
Trend study, obtains trend time series.Deep learning model in the embodiment of the present application can be deep learning shot and long term note
Recall model (LSTM) model or other neural network models and other can be used in the model of learning data rule.It is excellent
Selection of land, the embodiment of the present application selected depth Chief Learning Officer, CLO's short-term memory model (LSTM) model.
For example, excavating time series using LSTM model to 2 months -2017 years 2 months 2015 history waybill data with anti-
Reflect the cycle variation law and Near-future Development Trend rule of data.As shown in figure 4, history waybill data are input to LSTM model
In, it extracts history cycle time series and carries out polymerization as cycle information.History waybill data are input to LSTM model, are extracted
Historical trending time sequence is as tendency information.
Wherein, history cycle time series is extracted using history waybill data, as shown in figure 5, if extracting history waybill
The time series on some Monday temporally recalls mode on the basis of the Monday in data, before extracting the Monday
12 weeks Monday data are separately input to be handled in LSTM module, after 64 layers of full articulamentum processing, are output to
32 layers of full articulamentum processing, to generate the time series on the Monday.
The embodiment of the present application introduces external data in data processing stage further to analyze the variation of wave time part amount.It is right
External data is identified processing, obtains second processing result.
There are many modes for the acquisition of external data.For example, being extracted from internal data, or obtained by external data resource
It takes.It to the external data of acquisition, needs to be further processed, is converted into the feature for influencing wave time part amount.The application is implemented
The external data extracted in example can be technical dates data and weather data.Wherein, technical dates includes that the legal section in part is false
Day and part technical dates, for example, New Year's Day, the Spring Festival, the Dragon Boat Festival, Clear and Bright, May Day, mid-autumn, National Day, double 11, double ten second-class.To spy
The processing of different date data, it can be understood as with New Year's Day, the Spring Festival, the Dragon Boat Festival, Clear and Bright, May Day, mid-autumn, National Day, double 11, double 12
For column, date data is row, and the value range of data is [- 15,16] in each column.It is with the Spring Festival (i.e. the lunar New Year's Day of the lunar calendar)
Example, use 16 indicate 15 days before and after the date be not the Spring Festival, -15 indicate that there are also 15 days be the Spring Festival, and on the day of 0 indicates the Spring Festival, 15 are indicated
The Spring Festival passes by 15 days.By being identified processing to technical dates, technical dates and wave time part amount can be preferably excavated
Between influence relationship.
Processing for weather data, it can be understood as the case where indicating weather with numerical value, such as indicate the numerical value of weather
Range is { -2,0,2 }, wherein -2 indicate exceedingly odious weather, 2 indicate sunny, mends 0 when uncertain or normal.It is right
The processing of weather data, in order to preferably analyze influence of the weather conditions to wave time part amount.
After handling technical dates data and weather data, polymerization generates external information.
Optionally, data processing unit 801 can also include:
First processing subelement 8011, for extracting characteristic attribute data from history waybill data, and to characteristic attribute
Data are handled, and the first processing result is obtained；
First time sequence generates subelement 8012, for carrying out the period to history waybill data based on deep learning model
Rule study obtains multiple first time sequences, and is cycle time sequence by multiple first time sequential polymerizations；
Second time series generates subelement 8013, recent for being carried out based on deep learning model to history waybill data
Trend study, obtains trend time series；
Second processing subelement 8014 is identified processing to external data, obtains second processing as a result, said external number
According to including technical dates data and/or weather data.
Basic model unit 802 is created, is trained using data processed result, obtains depth residual error network foundation mould
Type.
Data processed result is obtained after handling history part amount data, data processed result polymerization is handled, is used
In training depth residual error network foundation model.
In the embodiment of the present application, by the first processing result, cycle time sequence, trend time series and second processing result
Polymerization, obtains polymerization result.Polymerization result is input to depth residual error network model, is trained.The depth residual error network mould
Type may include multiple residual error modules.As shown in fig. 6, by the first processing result, cycle time sequence, trend time series and
The polymerization of two processing results is input to depth residual error network foundation model.The depth residual error network foundation model includes 3 residual error moulds
Block.Each residual error module is built-up by a full articulamentum progress residual error study.
Optionally, basic model unit 802 is created, comprising:
It polymerize subelement 8021, is used for aggregated data processing result, obtains polymerization result；
Training subelement 8022 obtains depth residual error network foundation model for being trained using polymerization result.
Wherein polymerization subelement 8021 is also used to processing result, multiple first time sequences, the second time series and the
The polymerization of two processing results, obtains polymerization result.
Predicting unit 803 is updated, for updating aforementioned base model using the history waybill data in the second time range,
Prediction model is obtained to determine the prediction part amount of target wave time.
The time of the existing each every wave in site time of time series models training is longer, and generally higher than 20 minutes, daily
About 30 waves of wave time.As it can be seen that the training time that existing time series models expend is too long, lead to the forecasting efficiency of model
It is lower.For the variation for more effectively predicting non-incoming wave time part amount, need to construct the better model of more new capability, for promoting work
Efficiency.
The embodiment of the present application constructs depth residual error network foundation model, to the basic model, Ke Yitong in step 202
The mode of setting rolling forecast is crossed, to update the basic model.
For example, updating the basic model using the history waybill data in the second time range.Second time range is set
Setting can adjust according to data more new state dynamic.For example, prediction target is in March, 2017, then the second time range can be set
It is set to 2 months -2017 years in October, 2016.If prediction target is in April, 2017, the second time range can be set to 2016
In March, -2017 in November year.Alternatively, further reducing time range, such as predict that target is in March, 2017, then the second time model
It encloses and can be set to 2 months -2017 years in December, 2016.
The selection of the second time range is the mode monthly rolled in the embodiment of the present application, chooses the model of setting time span
It encloses.The time span can be the integral multiple of calendar month.For example, 3 months or 5 months.
After the history waybill data for obtaining the second time range, by the history waybill data further division of this period
At training dataset and validation data set.And re -training is carried out to basic model to update basic mould using training dataset
Type inputs updated basic model based on validation data set to obtain the wave of prediction time part amount, by dividing training dataset
With the mode of validation data set cross validation, to prevent updated model from overfitting problem occur.
Predicting unit 803 is updated, optionally, comprising:
Determine subelement 8031, the history waybill data of the second time range for determining prediction target wave time；
Data divide subelement 8032, for by the history waybill data of the second time range be divided into training dataset and
Validation data set；
Subelement 8033 is updated, for utilizing training dataset re -training depth residual error network foundation model, is obtained pre-
Survey model；
It predicts subelement 8034, for validation data set to be input to prediction model, obtains the prediction part amount of target wave time.
For example, the second time range is 5 months by taking the wave time part amount for predicting in March, 2017 as an example, that is, choose 2016 10
The history waybill data of 2 months -2017 years moon.History waybill data are divided into training dataset and validation data set.
For example, { in January, -2017 in October, 2016 } is training dataset, { 2 months 2017 } are validation data set；
Basic model is trained with training dataset, the basic model updated, indicates referred to as pre- for difference
Survey model；Validation data set is input to prediction model, obtains prediction result, as the prediction part amount of target wave time.
Optionally, the embodiment of the present application by adjusting training set and is tested in order to further enhance the stability of prediction result
The multiple basic models of training for demonstrate,proving data set carry out cross validation to realize the prediction of wave time part amount.
Optionally, data divide subelement 8032, are also used in the history waybill data of the second time range, dynamically
Multiple validation data sets are specified, and mark off the multiple and one-to-one training dataset of the validation data set.
Optionally, subelement 8033 is updated, is also used to inputting multiple training datasets into depth residual error network foundation respectively
Model re -training obtains multiple and training dataset and corresponds depth residual error Network Prediction Model.
Optionally, subelement 8034 is predicted, further includes:
Subelement is predicted, for that will input respectively with the one-to-one validation data set of multiple training datasets and multiple instructions
Practice the one-to-one depth residual error network model of data set, obtains multiple prediction part amounts；
Computation subunit, the prediction part amount for calculating the average value of multiple prediction part amounts, as target wave time.
For example, the second time range is 5 months by taking the wave time part amount for predicting in March, 2017 as an example, that is, choose 2016 10
The history waybill data of 2 months -2017 years moon.
The history waybill data in 2 months -2017 years in October, 2016 are divided into 3 kinds of different training datasets and verifying number
According to collection.
For example, { in January, -2017 in October, 2016 } is the first training dataset, { 2 months 2017 } are first verification data
Collection；
{ in October, -2016 in December, 2016,2 months 2017 } is the second training dataset, and { in January, 2017 } tests for second
Demonstrate,prove data set；
{ October in November, 2016-,-2 months in January, 2017 } is third training dataset, and { in December, 2016 } tests for third
Demonstrate,prove data set.
Basic model is trained with the first training dataset, the basic model updated, indicates to be referred to as difference
For the first prediction model；First verification data collection is input to the first prediction model, obtains the first prediction result.
Basic model is trained with the second training dataset, the basic model updated, indicates to be referred to as difference
For the second prediction model；First verification data collection is input to the second prediction model, obtains the second prediction result.
Basic model is trained with third training dataset, the basic model updated, indicates to be referred to as difference
For third prediction model；First verification data collection is input to the second prediction model, obtains third prediction result.
Finally, the first prediction result, the second prediction result and the second prediction result are averaged, final prediction is obtained
As a result.
It should be appreciated that in the method that all units or module recorded in device 700 or 800 and reference Fig. 1 or Fig. 2 are described
Each step is corresponding.Device 700 or 800 and wherein is equally applicable to above with respect to the operation and feature of method description as a result,
The unit for including, details are not described herein.Device 700 or 800 can be realized in advance in the browser of electronic equipment or other safety
In, it can also be loaded into the browser or its security application of electronic equipment by modes such as downloadings.Device 700 or
Corresponding units in 800 can be cooperated with the unit in electronic equipment to realize the scheme of the embodiment of the present application.
Below with reference to Fig. 9, it illustrates the calculating of the terminal device or server that are suitable for being used to realize the embodiment of the present application
The structural schematic diagram of machine system 900.
As shown in figure 9, computer system 900 includes central processing unit (CPU) 901, it can be read-only according to being stored in
Program in memory (ROM) 902 or be loaded into the program in random access storage device (RAM) 903 from storage section 908 and
Execute various movements appropriate and processing.In RAM 903, also it is stored with system 900 and operates required various programs and data.
CPU 901, ROM 902 and RAM 903 are connected with each other by bus 904.Input/output (I/O) interface 905 is also connected to always
Line 904.
I/O interface 905 is connected to lower component: the importation 906 including keyboard, mouse etc.；It is penetrated including such as cathode
The output par, c 907 of spool (CRT), liquid crystal display (LCD) etc. and loudspeaker etc.；Storage section 908 including hard disk etc.；
And the communications portion 909 of the network interface card including LAN card, modem etc..Communications portion 909 via such as because
The network of spy's net executes communication process.Driver 910 is also connected to I/O interface 905 as needed.Detachable media 911, such as
Disk, CD, magneto-optic disk, semiconductor memory etc. are mounted on as needed on driver 910, in order to read from thereon
Computer program be mounted into storage section 908 as needed.
Particularly, in accordance with an embodiment of the present disclosure, it may be implemented as computer above with reference to the process described of Fig. 1 or 2
Software program.For example, embodiment of the disclosure includes a kind of computer program product comprising be tangibly embodied in machine readable
Computer program on medium, the computer program include the program code for executing the method for Fig. 1 or 2.Such
In embodiment, which can be downloaded and installed from network by communications portion 909, and/or is situated between from detachable
Matter 911 is mounted.
Flow chart and block diagram in attached drawing are illustrated according to the system of various embodiments of the invention, method and computer journey
The architecture, function and operation in the cards of sequence product.In this regard, each box in flowchart or block diagram can generation
A part of one module, program segment or code of table, a part of the module, program segment or code include one or more
Executable instruction for implementing the specified logical function.It should also be noted that in some implementations as replacements, institute in box
The function of mark can also occur in a different order than that indicated in the drawings.For example, two boxes succeedingly indicated are practical
On can be basically executed in parallel, they can also be executed in the opposite order sometimes, and this depends on the function involved.Also it wants
It is noted that the combination of each box in block diagram and or flow chart and the box in block diagram and or flow chart, Ke Yiyong
The dedicated hardware based system of defined functions or operations is executed to realize, or can be referred to specialized hardware and computer
The combination of order is realized.
Being described in the embodiment of the present application involved unit or module can be realized by way of software, can also be with
It is realized by way of hardware.Described unit or module also can be set in the processor, for example, can be described as:
A kind of processor includes data processing unit, creation basic model unit and updates predicting unit.Wherein, these units or mould
The title of block does not constitute the restriction to the unit or module itself under certain conditions, for example, data processing unit can be with
It is described as " for handling the unit of data ".
As on the other hand, present invention also provides a kind of computer readable storage medium, the computer-readable storage mediums
Matter can be computer readable storage medium included in device described in above-described embodiment；It is also possible to individualism, not
The computer readable storage medium being fitted into equipment.Computer-readable recording medium storage has one or more than one journey
Sequence, described program are used to execute the part amount prediction technique for being described in the application by one or more than one processor.
Above description is only the preferred embodiment of the application and the explanation to institute's application technology principle.Those skilled in the art
Member is it should be appreciated that invention scope involved in the application, however it is not limited to technology made of the specific combination of above-mentioned technical characteristic
Scheme, while should also cover in the case where not departing from the inventive concept, it is carried out by above-mentioned technical characteristic or its equivalent feature
Any combination and the other technical solutions formed.Such as features described above has similar function with (but being not limited to) disclosed herein
Can technical characteristic replaced mutually and the technical solution that is formed.
Claims (20)
1. a kind of amount prediction technique, which is characterized in that the described method includes:
To within the scope of first time history waybill data and external data handle, obtain data processed result；
It is trained using the data processed result, obtains depth residual error network foundation model；
The basic model is updated using the history waybill data in the second time range, obtains prediction model to determine target wave
Secondary prediction part amount.
2. the method according to claim 1, wherein the history waybill data within the scope of first time and
External data is handled, and data processed result is obtained, comprising:
Characteristic attribute data are extracted from the history waybill data, and the characteristic attribute data are handled, and obtain the
One processing result；
Periodic law study is carried out to the history waybill data based on deep learning model, obtains multiple first time sequences,
And by the multiple first time sequential polymerization be cycle time sequence；
Recent trend study is carried out to the history waybill data based on the deep learning model, obtains trend time series；
Processing is identified to the external data, obtains second processing as a result, the external data includes technical dates data
And/or weather data.
3. method according to claim 1 or 2, which is characterized in that it is described to be trained using the data processed result,
Obtain depth residual error network foundation model, comprising:
It polymerize the data processed result, obtains polymerization result；
It is trained using the polymerization result, obtains depth residual error network foundation model.
4. according to the method described in claim 3, it is characterized in that, the polymerization data processed result, obtains polymerization knot
Fruit, comprising:
First processing result, the cycle time sequence, the trend time series and the second processing are gathered
It closes, obtains polymerization result.
5. method according to claim 1-4, which is characterized in that the history using in the second time range
Waybill data update the basic model, obtain prediction model to determine the prediction part amount of target wave time, comprising:
Determine the history waybill data of the second time range of prediction target wave time；
The history waybill data of second time range are divided into training dataset and validation data set；
Using depth residual error network foundation model described in the training dataset re -training, prediction model is obtained；
The validation data set is input to the prediction model, obtains the prediction part amount of target wave time.
6. according to the method described in claim 5, it is characterized in that, the history waybill data by the second time range divide
For training dataset and validation data set, comprising:
In the history waybill data of second time range, multiple first verification data collection are dynamically specified, and draw
Separate multiple and one-to-one first training dataset of the first verification data collection.
7. according to the method described in claim 6, it is characterized in that, described using deep described in the training dataset re -training
Residual error network foundation model is spent, prediction model is obtained, comprising:
The multiple first training dataset is inputted into the depth residual error network foundation model re -training respectively, is obtained multiple
Depth residual error Network Prediction Model is corresponded with first training dataset.
8. the method according to the description of claim 7 is characterized in that described be input to the prediction mould for the validation data set
Type obtains the prediction part amount of target wave time, comprising:
Will with the multiple first training dataset correspondingly the first verification data collection input respectively with it is the multiple
The one-to-one first depth residual error network model of first training dataset obtains multiple first prediction part amounts；
Calculate the average value of the multiple first prediction part amount, the prediction part amount as the target wave time.
9. according to the described in any item methods of claim 2-8, which is characterized in that described to be extracted from the history waybill data
Characteristic attribute data, and the characteristic attribute data are handled, comprising:
Characteristic attribute data are extracted from the history waybill data, the characteristic attribute data include at least: number node number
According to, month data, date data, week data, wave time part amount data；
Dimension-reduction treatment is carried out to the characteristic attribute data.
10. according to the method described in claim 9, it is characterized in that, it is described to the characteristic attribute data carry out dimension-reduction treatment,
Include:
The characteristic attribute data are mapped to by feature space using embedding mode.
11. a kind of amount prediction meanss, described device include:
Data processing unit, within the scope of first time history waybill data and external data handle, counted
According to processing result；
It creates basic model unit and obtains depth residual error network foundation mould for being trained using the data processed result
Type；
Predicting unit is updated, for updating the basic model using the history waybill data in the second time range, is obtained pre-
Model is surveyed to determine the prediction part amount of target wave time.
12. device according to claim 11, which is characterized in that the data processing unit, comprising:
First processing subelement, for extracting characteristic attribute data from the history waybill data, and to the characteristic attribute
Data are handled, and the first processing result is obtained；
First time sequence generates subelement, for carrying out periodic law to the history waybill data based on deep learning model
Study obtains multiple first time sequences, and is cycle time sequence by the multiple first time sequential polymerization；
Second time series generates subelement, recent for being carried out based on the deep learning model to the history waybill data
Trend study, obtains trend time series；
Second processing subelement is identified processing to the external data, obtains second processing as a result, the external data packet
Include technical dates data and/or weather data.
13. device according to claim 11 or 12, which is characterized in that the creation basic model unit, comprising:
It polymerize subelement and obtains polymerization result for polymerizeing the data processed result；
Training subelement obtains depth residual error network foundation model for being trained using the polymerization result.
14. device according to claim 13, which is characterized in that the polymerized unit is also used to handle described first
As a result, the cycle time sequence, the trend time series and the second processing result are polymerize, and obtain polymerization knot
Fruit.
15. the described in any item devices of 1-14 according to claim 1, which is characterized in that the update predicting unit, comprising:
Determine subelement, the history waybill data of the second time range for determining prediction target wave time；
Data divide subelement, for the history waybill data of the second time range to be divided into training dataset and verify data
Collection；
Subelement is updated, for obtaining pre- using depth residual error network foundation model described in the training dataset re -training
Survey model；
It predicts subelement, for the validation data set to be input to the prediction model, obtains the prediction part amount of target wave time.
16. device according to claim 15, which is characterized in that the data divide subelement, are also used to described the
In the history waybill data of two time ranges, multiple first verification data collection are dynamically specified, and mark off multiple and institute
State one-to-one first training dataset of first verification data collection.
17. device according to claim 16, which is characterized in that the update subelement is also used to the multiple
One training dataset inputs the depth residual error network foundation model re -training respectively, obtains the multiple and described first training number
Depth residual error Network Prediction Model is corresponded according to collection.
18. device according to claim 17, which is characterized in that the prediction subelement, comprising:
First prediction subelement, for will be with the multiple first training dataset first verification data collection correspondingly
Input and the one-to-one first depth residual error network model of the multiple first training dataset respectively, it is pre- to obtain multiple first
Survey part amount；
Computation subunit, the prediction part for calculating the average value of the multiple first prediction part amount, as the target wave time
Amount.
19. a kind of equipment, including processor, storage device；It is characterized by:
The storage device, for storing one or more programs；
When one or more of programs are executed by the processor, so that the processor is realized as appointed in claim 1-10
Method described in one.
20. a kind of computer readable storage medium, is stored thereon with computer program, the computer program is executed by processor
When, realize such as method of any of claims 1-10.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711426164.0A CN110019401B (en) | 2017-12-25 | 2017-12-25 | Method, device, equipment and storage medium for predicting part quantity |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711426164.0A CN110019401B (en) | 2017-12-25 | 2017-12-25 | Method, device, equipment and storage medium for predicting part quantity |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110019401A true CN110019401A (en) | 2019-07-16 |
| CN110019401B CN110019401B (en) | 2024-04-05 |
Family
ID=67187147
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711426164.0A Active CN110019401B (en) | 2017-12-25 | 2017-12-25 | Method, device, equipment and storage medium for predicting part quantity |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110019401B (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110889560A (en) * | 2019-12-06 | 2020-03-17 | 西北工业大学 | Express delivery sequence prediction method with deep interpretability |
| CN111401940A (en) * | 2020-03-05 | 2020-07-10 | 杭州网易再顾科技有限公司 | Feature prediction method, feature prediction device, electronic device, and storage medium |
| CN112183832A (en) * | 2020-09-17 | 2021-01-05 | 上海东普信息科技有限公司 | Express pickup quantity prediction method, device, equipment and storage medium |
| CN112418534A (en) * | 2020-11-26 | 2021-02-26 | 上海东普信息科技有限公司 | Method and device for predicting collection quantity, electronic equipment and computer readable storage medium |
| CN112801327A (en) * | 2019-11-14 | 2021-05-14 | 顺丰科技有限公司 | Method, device, equipment and storage medium for predicting logistics flow and modeling thereof |
| CN112862137A (en) * | 2019-11-27 | 2021-05-28 | 顺丰科技有限公司 | Method and device for predicting quantity, computer equipment and computer readable storage medium |
| CN112907267A (en) * | 2019-12-03 | 2021-06-04 | 顺丰科技有限公司 | Method and device for predicting cargo quantity, computer equipment and storage medium |
| CN112948763A (en) * | 2019-12-11 | 2021-06-11 | 顺丰科技有限公司 | Method and device for predicting quantity of component, electronic equipment and storage medium |
| CN112966849A (en) * | 2019-12-13 | 2021-06-15 | 顺丰科技有限公司 | Method, device and equipment for establishing component prediction model |
| CN112990526A (en) * | 2019-12-16 | 2021-06-18 | 顺丰科技有限公司 | Method and device for predicting logistics arrival quantity and storage medium |
| CN112990520A (en) * | 2019-12-13 | 2021-06-18 | 顺丰科技有限公司 | Mesh point connection quantity prediction method and device, computer equipment and storage medium |
| CN113077069A (en) * | 2020-01-03 | 2021-07-06 | 顺丰科技有限公司 | Modeling method, device and equipment for predicting shift quantity and storage medium |
| US12265942B2 (en) | 2021-05-17 | 2025-04-01 | Starbucks Corporation | Method and system for automatic replenishment of retail enterprise store, and computer-readable storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105427194A (en) * | 2015-12-21 | 2016-03-23 | 西安美林数据技术股份有限公司 | Method and device for electricity sales amount prediction based on random forest regression |
| CN106886846A (en) * | 2017-04-26 | 2017-06-23 | 中南大学 | A kind of bank outlets' excess reserve Forecasting Methodology that Recognition with Recurrent Neural Network is remembered based on shot and long term |
| US20170249534A1 (en) * | 2016-02-29 | 2017-08-31 | Fujitsu Limited | Method and apparatus for generating time series data sets for predictive analysis |
| CN107316083A (en) * | 2017-07-04 | 2017-11-03 | 北京百度网讯科技有限公司 | Method and apparatus for updating deep learning model |
| CN107330522A (en) * | 2017-07-04 | 2017-11-07 | 北京百度网讯科技有限公司 | Method, apparatus and system for updating deep learning model |
-
2017
- 2017-12-25 CN CN201711426164.0A patent/CN110019401B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105427194A (en) * | 2015-12-21 | 2016-03-23 | 西安美林数据技术股份有限公司 | Method and device for electricity sales amount prediction based on random forest regression |
| US20170249534A1 (en) * | 2016-02-29 | 2017-08-31 | Fujitsu Limited | Method and apparatus for generating time series data sets for predictive analysis |
| CN106886846A (en) * | 2017-04-26 | 2017-06-23 | 中南大学 | A kind of bank outlets' excess reserve Forecasting Methodology that Recognition with Recurrent Neural Network is remembered based on shot and long term |
| CN107316083A (en) * | 2017-07-04 | 2017-11-03 | 北京百度网讯科技有限公司 | Method and apparatus for updating deep learning model |
| CN107330522A (en) * | 2017-07-04 | 2017-11-07 | 北京百度网讯科技有限公司 | Method, apparatus and system for updating deep learning model |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112801327A (en) * | 2019-11-14 | 2021-05-14 | 顺丰科技有限公司 | Method, device, equipment and storage medium for predicting logistics flow and modeling thereof |
| CN112862137A (en) * | 2019-11-27 | 2021-05-28 | 顺丰科技有限公司 | Method and device for predicting quantity, computer equipment and computer readable storage medium |
| CN112907267B (en) * | 2019-12-03 | 2024-08-23 | 顺丰科技有限公司 | Method, apparatus, computer device and storage medium for predicting article quantity |
| CN112907267A (en) * | 2019-12-03 | 2021-06-04 | 顺丰科技有限公司 | Method and device for predicting cargo quantity, computer equipment and storage medium |
| CN110889560A (en) * | 2019-12-06 | 2020-03-17 | 西北工业大学 | Express delivery sequence prediction method with deep interpretability |
| CN112948763A (en) * | 2019-12-11 | 2021-06-11 | 顺丰科技有限公司 | Method and device for predicting quantity of component, electronic equipment and storage medium |
| CN112948763B (en) * | 2019-12-11 | 2024-04-09 | 顺丰科技有限公司 | Piece quantity prediction method and device, electronic equipment and storage medium |
| CN112966849A (en) * | 2019-12-13 | 2021-06-15 | 顺丰科技有限公司 | Method, device and equipment for establishing component prediction model |
| CN112990520A (en) * | 2019-12-13 | 2021-06-18 | 顺丰科技有限公司 | Mesh point connection quantity prediction method and device, computer equipment and storage medium |
| CN112966849B (en) * | 2019-12-13 | 2024-06-07 | 顺丰科技有限公司 | Method, device and equipment for establishing part quantity prediction model |
| CN112990520B (en) * | 2019-12-13 | 2024-08-20 | 顺丰科技有限公司 | Method, device, computer equipment and storage medium for predicting net point connection piece quantity |
| CN112990526A (en) * | 2019-12-16 | 2021-06-18 | 顺丰科技有限公司 | Method and device for predicting logistics arrival quantity and storage medium |
| CN113077069A (en) * | 2020-01-03 | 2021-07-06 | 顺丰科技有限公司 | Modeling method, device and equipment for predicting shift quantity and storage medium |
| CN113077069B (en) * | 2020-01-03 | 2023-06-13 | 顺丰科技有限公司 | Modeling method, device, equipment and storage medium for predicting shift number |
| CN111401940A (en) * | 2020-03-05 | 2020-07-10 | 杭州网易再顾科技有限公司 | Feature prediction method, feature prediction device, electronic device, and storage medium |
| CN112183832A (en) * | 2020-09-17 | 2021-01-05 | 上海东普信息科技有限公司 | Express pickup quantity prediction method, device, equipment and storage medium |
| CN112418534A (en) * | 2020-11-26 | 2021-02-26 | 上海东普信息科技有限公司 | Method and device for predicting collection quantity, electronic equipment and computer readable storage medium |
| CN112418534B (en) * | 2020-11-26 | 2022-10-14 | 上海东普信息科技有限公司 | Method and device for predicting quantity of collected parts, electronic equipment and computer readable storage medium |
| US12265942B2 (en) | 2021-05-17 | 2025-04-01 | Starbucks Corporation | Method and system for automatic replenishment of retail enterprise store, and computer-readable storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110019401B (en) | 2024-04-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110019401A (en) | Part amount prediction technique, device, equipment and its storage medium | |
| Holdaway | Harness oil and gas big data with analytics: Optimize exploration and production with data-driven models | |
| Suryani et al. | Demand scenario analysis and planned capacity expansion: A system dynamics framework | |
| CN110400021A (en) | Bank outlets' cash dosage prediction technique and device | |
| Vadisetty | AI-Based Smart Governance | |
| CN109961248A (en) | Waybill complains prediction technique, device, equipment and its storage medium | |
| CN109961165A (en) | Part amount prediction technique, device, equipment and its storage medium | |
| Sun et al. | Multi-step ahead tourism demand forecasting: The perspective of the learning using privileged information paradigm | |
| CN111738504A (en) | Enterprise financial index fund amount prediction method and device, equipment and storage medium | |
| US20200050982A1 (en) | Method and System for Predictive Modeling for Dynamically Scheduling Resource Allocation | |
| CN112380321B (en) | Primary and secondary database allocation method based on bill knowledge graph and related equipment | |
| CN110263136B (en) | Method and device for pushing object to user based on reinforcement learning model | |
| EP4433938A1 (en) | Optimization and decision-making using causal aware machine learning models trained from simulators | |
| Hu et al. | Generating decision rules for flexible capacity expansion problem through gene expression programming | |
| CN113793037A (en) | Service distribution method, device, equipment and storage medium based on data analysis | |
| Pei et al. | A Predictive Analysis of the Business Environment of Economies along the Belt and Road Using the Fractional‐Order Grey Model | |
| CN111738508A (en) | Bank branch blank certificate inventory usage prediction method and device, equipment and medium | |
| Mutanov et al. | Investments decision making on the basis of system dynamics | |
| CN109979122A (en) | ATM safeguards method for early warning and system | |
| CN109767031A (en) | Model classifiers method for building up, device, computer equipment and storage medium | |
| CN115630979A (en) | Day-ahead electricity price prediction method and device, storage medium and computer equipment | |
| CN113298377A (en) | Method and device for screening items in enterprise research and development expense and deduction | |
| Blok et al. | Dynamic models of the firm: determining optimal investment, financing and production policies by computer | |
| Meng et al. | Application of EEMD+ BI_GRU hybrid model for intelligent service area traffic flow forecasting. | |
| Yang et al. | A Multitime Window Parallel Scheduling System for Large‐Scale Offshore Platform Project |
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 |