[go: up one dir, main page]

CN114841316B - Abnormal trajectory detection method and system based on recurrent neural network and differential autoencoder - Google Patents

Abnormal trajectory detection method and system based on recurrent neural network and differential autoencoder Download PDF

Info

Publication number
CN114841316B
CN114841316B CN202210439506.7A CN202210439506A CN114841316B CN 114841316 B CN114841316 B CN 114841316B CN 202210439506 A CN202210439506 A CN 202210439506A CN 114841316 B CN114841316 B CN 114841316B
Authority
CN
China
Prior art keywords
track
abnormal
data
neural network
encoder
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
Application number
CN202210439506.7A
Other languages
Chinese (zh)
Other versions
CN114841316A (en
Inventor
刘晓明
赵清源
沈超
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xian Jiaotong University
Original Assignee
Xian Jiaotong University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Xian Jiaotong University filed Critical Xian Jiaotong University
Priority to CN202210439506.7A priority Critical patent/CN114841316B/en
Publication of CN114841316A publication Critical patent/CN114841316A/en
Application granted granted Critical
Publication of CN114841316B publication Critical patent/CN114841316B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S19/00—Satellite radio beacon positioning systems; Determining position, velocity or attitude using signals transmitted by such systems
    • G01S19/38—Determining a navigation solution using signals transmitted by a satellite radio beacon positioning system
    • G01S19/39—Determining a navigation solution using signals transmitted by a satellite radio beacon positioning system the satellite radio beacon positioning system transmitting time-stamped messages, e.g. GPS [Global Positioning System], GLONASS [Global Orbiting Navigation Satellite System] or GALILEO
    • G01S19/393—Trajectory determination or predictive tracking, e.g. Kalman filtering
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/04—Inference or reasoning models
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Remote Sensing (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Train Traffic Observation, Control, And Security (AREA)

Abstract

The invention discloses an abnormal track detection method and system based on a cyclic neural network and a differential self-encoder, wherein the method comprises the steps of segmenting acquired original track data to acquire segmented track data; dividing a target map into a plurality of grids, carrying out discretization processing on the segmented track data to obtain discretized track data, dividing the discretized track data to obtain a training set and a testing set, constructing an abnormal track detection model based on a cyclic neural network and a differential self-encoder, inputting the training set into the abnormal track detection model for training to obtain an optimized abnormal track detection model, inputting the testing set into the optimized abnormal track detection model to obtain an abnormal score of the track, and judging whether the track is abnormal or not based on the abnormal score. The method can code the track information of the vehicle through the generation model, explore the expression form of the track anomaly in the potential space, judge the anomaly and improve the detection accuracy.

Description

Abnormal track detection method and system based on cyclic neural network and differential self-encoder
Technical Field
The invention belongs to the technical field of track detection, and relates to an abnormal track detection method and system based on a cyclic neural network and a differential self-encoder.
Background
The rapid development of the number of steps of global satellite positioning navigation systems (GPS) and handheld mobile terminals has prompted the progress of research based on mobile object location information. As portability increases and costs decrease, trajectory data of a large number of moving objects is collected and stored. There is a strong temporal, spatial correlation between these data, and these data with spatiotemporal properties characterize the movement characteristics of the object. While implicit local patterns tend to be complex, it is difficult to mine valid information from the data. Therefore, an effective data management mechanism needs to be established, a data mining technology in big data is researched, and valuable effective information such as space-time patterns is extracted from the rule data of a large number of moving objects.
Existing studies on abnormal trajectory detection can be divided into two categories, metric-based methods and learning-based methods. The first category is typically based on manual features, and since the trajectories between different sites are typically very different, these definitions may only be valid for a few cases, but not for many others. Second, track comparison is often time consuming and does not support efficient on-line detection. Many deep learning methods have been applied to anomaly detection problems, however, they lack a means to mine useful information about track anomalies in the underlying embedding space, i.e., they can only embed tracks into the low-dimensional space without exploring which part of the underlying space represents normals and which part represents anomalies. Thus, modeling an abnormal trajectory using a probabilistic model provides new ideas.
Disclosure of Invention
The invention aims to solve the problems in the prior art and provides an abnormal track detection method and system based on a cyclic neural network and a differential self-encoder, which can effectively improve the accuracy of abnormal track detection.
In order to achieve the purpose, the invention is realized by adopting the following technical scheme:
The abnormal track detection method based on the cyclic neural network and the differential self-encoder comprises the following steps:
segmenting the acquired original track data to acquire segmented track data;
dividing a target map into a plurality of grids, and discretizing the segmented track data to obtain discretized track data;
dividing discretized track data to obtain a training set and a testing set;
Constructing an abnormal track detection model based on the cyclic neural network and the differential self-encoder;
Inputting the training set into an abnormal track detection model for training to obtain an optimized abnormal track detection model;
Inputting the test set into an optimized abnormal track detection model to obtain the abnormal score of the track;
Based on the anomaly score, it is determined whether the trajectory is anomalous.
The invention further improves that:
segmenting the acquired original track data to acquire segmented track data, wherein the segmented track data comprises the following specific steps:
The method comprises the steps of acquiring track point position information and a time stamp of a vehicle in running track data, wherein based on the track point position information and the time stamp of the vehicle, a sequence P= { P 1,p2,…,pn } represents one track in the track data, wherein P i＝(ti,li) epsilon P represents the time stamp and the position information of an ith track point in the sequence P, and n is the length of the track;
based on a track segmentation rule, original track data P is segmented, a sequence Q= { Q 1,q2,…,qm } is used for representing one track in track data, wherein Q i＝(ti,li) epsilon Q represents a time stamp and position information of an ith track point in the sequence Q, and m is the length of the track.
The track segmentation rule specifically comprises the following steps:
When the space deviation of two adjacent track points is larger than a set first threshold value, lagrange interpolation is carried out through surrounding track points to form complete track data, or when the track points do not displace within a set time or the displacement deviation of the track points does not reach a set second threshold value, the track is segmented to form complete track data;
The track is divided into the following specific steps:
and eliminating the track points which do not displace within the set time, and dividing the displacement points which do not move to obtain segmented track data.
Dividing the target map into a plurality of grids, discretizing the segmented track data, and obtaining discretized track data, wherein the discretized track data comprises the following concrete steps of:
The method comprises the steps of dividing a target map into a plurality of grids, enabling track points falling into the same grids to have the same numbers, achieving discretization of track data points, using a sequence R= { R 1,r2,…,rm } to represent one track in track data, wherein R i＝(ti,li) epsilon R represents time stamps and position information of ith track points in the sequence R, and m is the length of the track.
Based on a cyclic neural network and a differential self-encoder, an abnormal track detection model is constructed, specifically:
The track self-encoder model is built, and comprises an encoder E 1, a probability distribution P 1 and a generator G 1, wherein the encoder E 1 is a cyclic neural network, track data R 1 in a training set is taken as input, potential representation R R of a track is output, the probability distribution P 1 of a path is a probability distribution model based on Gaussian distribution, the potential representation R R of the track is taken as input, mu= { mu 1,μ2,…,μn } is output to represent the type of the path in a potential space, n is the number of the types, the generator G 1 is a cyclic neural network, the potential representation R R of the track is taken as input, and the reconstructed track is output
Constructing a discriminator D which is a k-layer fully connected neural network, and combining the track data R 1 in the training set and the reconstructed trackAs input, output scoring matrix
Inputting the training set into the abnormal track detection model for training to obtain an optimized abnormal track detection model, wherein the method specifically comprises the following steps:
step 1, changing the lower bound by track information As a loss function, using trajectory data R 1 in the training set, the trajectory potentially representing μ, training the neural network parameters in encoder E 1 and generator G 1 of the user-variational automatic encoder using a random gradient descent method;
Wherein, Θ 1 represents the neural network parameters in the encoder E 1 and generator G 1, respectively, to be trained; representing a trajectory Probability of belonging to normal path;
And 2, training the abnormal track detection model in the step 1 until the loss function is lower than the set threshold value.
Inputting the test set into an optimized abnormal track detection model, and acquiring the abnormal score of the track, wherein the abnormal score is specifically as follows:
Inputting the trajectory data R 2 in the test set to the encoder section in the optimized abnormal trajectory detection model, the resulting output being a potential representation R R of the trajectory;
the probability of conforming to the path μ k εμ in the potential space is calculated based on the potential representation R R, and the anomaly score s (R) for the trace is calculated from the probability.
The method for calculating the anomaly score s (R) comprises the following steps:
Where μ c represents the path potential representation in the potential vector space.
An anomaly track detection system based on a recurrent neural network and a differential self-encoder, comprising:
The segmentation module is used for segmenting the acquired original track data to acquire segmented track data;
the first dividing module is used for dividing the target map into a plurality of grids, and discretizing the segmented track data to obtain discretized track data;
The second division module is used for dividing the discretized track data to obtain a training set and a testing set;
The construction module is used for constructing an abnormal track detection model based on the cyclic neural network and the differential self-encoder;
the training module is used for inputting the training set into the abnormal track detection model for training to obtain an optimized abnormal track detection model;
the testing module is used for inputting the testing set into the optimized abnormal track detection model to obtain the abnormal score of the track;
and the judging module is used for judging whether the track is abnormal or not based on the abnormal score.
Compared with the prior art, the invention has the following beneficial effects:
The method extracts track data of the vehicle, performs segmentation and discretization, builds and trains an abnormal track detection model at the same time, and judges whether the track is abnormal or not. The method can code the track information of the vehicle through the generation model, explore the expression form of the track anomaly in the potential space, judge the anomaly based on the probability model, and make full use of the known information in a training mode to enable the obtained track potential representation to contain more complete information, so as to obtain higher anomaly detection accuracy.
Drawings
For a clearer description of the technical solutions of the embodiments of the present invention, the drawings that are needed in the embodiments will be briefly described below, it being understood that the following drawings only illustrate some embodiments of the present invention and should not be considered as limiting the scope, and other related drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a method of anomaly track detection based on a recurrent neural network and a differential self-encoder;
FIG. 2 is a block diagram of an anomaly trajectory detection method based on a recurrent neural network and a differential self-encoder;
FIG. 3 is a path probability inference flow chart;
Fig. 4 is a block diagram of an abnormal track detection system based on a recurrent neural network and a differential self-encoder.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. The components of the embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.
Thus, the following detailed description of the embodiments of the invention, as presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
It should be noted that like reference numerals and letters refer to like items in the following figures, and thus once an item is defined in one figure, no further definition or explanation thereof is necessary in the following figures.
In the description of the embodiments of the present invention, it should be noted that, if the terms "upper," "lower," "horizontal," "inner," and the like indicate an azimuth or a positional relationship based on the azimuth or the positional relationship shown in the drawings, or the azimuth or the positional relationship in which the inventive product is conventionally put in use, it is merely for convenience of describing the present invention and simplifying the description, and does not indicate or imply that the apparatus or element to be referred to must have a specific azimuth, be configured and operated in a specific azimuth, and thus should not be construed as limiting the present invention. Furthermore, the terms "first," "second," and the like, are used merely to distinguish between descriptions and should not be construed as indicating or implying relative importance.
Furthermore, the term "horizontal" if present does not mean that the component is required to be absolutely horizontal, but may be slightly inclined. As "horizontal" merely means that its direction is more horizontal than "vertical", and does not mean that the structure must be perfectly horizontal, but may be slightly inclined.
In the description of the embodiments of the present invention, it should also be noted that, unless explicitly stated or limited otherwise, the terms "disposed," "mounted," "connected," and "connected" should be interpreted broadly, and may be, for example, fixedly connected, detachably connected, or integrally connected, mechanically connected, electrically connected, directly connected, indirectly connected through an intermediate medium, or communicating between two elements. The specific meaning of the above terms in the present invention can be understood by those of ordinary skill in the art according to the specific circumstances.
The invention is described in further detail below with reference to the attached drawing figures:
Referring to fig. 1 and 2, the invention discloses an abnormal track detection method based on a cyclic neural network and a differential self-encoder, which comprises the following steps:
S101, segmenting the acquired original track data to acquire segmented track data.
The method comprises the steps of acquiring track point position information and a time stamp of a vehicle in running track data, wherein based on the track point position information and the time stamp of the vehicle, a sequence P= { P 1,p2,…,pn } represents one track in the track data, wherein P i＝(ti,li) epsilon P represents the time stamp and the position information of an ith track point in the sequence P, and n is the length of the track;
based on a track segmentation rule, original track data P is segmented, a sequence Q= { Q 1,q2,…,qm } is used for representing one track in track data, wherein Q i＝(ti,li) epsilon Q represents a time stamp and position information of an ith track point in the sequence Q, and m is the length of the track.
In this embodiment, the track data set is a T-Drive trajectory data set, and one taxi driving track data set is derived from:
https:// www.microsoft.com/en-us/research/publication/t-drive-trajectory-data-sa mple, the data set comprising a circle of tracks of 10357 taxis, the total number of points in the data set being about 1500 ten thousand, the total distance of the tracks reaching 900 ten thousand kilometers. Basic information such as track points and time stamps of the vehicle is acquired in the running track data in a centralized manner.
A track in the track data is represented by a sequence P＝{p1,p2,…,pn}＝{(0,116.51,39.93),(600,116.51,39.93),(600,116.47,39.91),…,(1456,116.47,39.90)}, where P i＝(ti,li) ∈p represents the timestamp and position information of the ith track point in the sequence P, and n is the length of the track.
The original track data P is segmented according to the following segmentation rule, namely, when the space deviation of two adjacent track points is larger than a set first threshold value, lagrange interpolation is carried out on surrounding track points to form complete track data, or when the track points are not displaced within set time or the displacement deviation of the track points does not reach a set second threshold value, the track is segmented to form complete track data.
The track is divided into the following specific steps:
and eliminating the track points which do not displace within the set time, and dividing the displacement points which do not move to obtain segmented track data.
In the above sequence, the positions of p 1 and p 2 are unchanged, which means that the vehicle is not displaced within 600 seconds, so the track is divided, the track point of p 1 is removed, and a track in track data is represented by a sequence Q＝{q1,q2,…,qm}＝{(600,116.51,39.93),(600,116.47,39.91),…,(1456,116.47,39.90)}, wherein Q i＝(ti,li) e Q represents the timestamp and position information of the ith track point in the sequence Q, and m is the length of the track.
S102, dividing a target map into a plurality of grids, and discretizing the segmented track data to obtain discretized track data;
The method comprises the steps of dividing a target map into a plurality of grids, enabling track points falling into the same grids to have the same numbers, achieving discretization of track data points, using a sequence R= { R 1,r2,…,rm } to represent one track in track data, wherein R i＝(ti,li) epsilon R represents time stamps and position information of ith track points in the sequence R, and m is the length of the track.
The whole map is divided into 300m grids, track points falling into the same grids have the same numbers, and discretization of the track data points is achieved. And selecting 200 grid points with the largest passing route as a departure point or a destination, and further dividing the track. The sequence r= { R 1,r2,…,rm } = { (600,184,35), (600,169,28), (1456,169,24) } is used to represent one track in the track data, where R i＝(ti,li) ∈r represents the timestamp and position information of the i-th track point in the sequence R, and m is the length of the track.
S103, dividing the discretized track data to obtain a training set and a testing set.
S104, constructing an abnormal track detection model based on the cyclic neural network and the differential self-encoder.
The track self-encoder model is built, and comprises an encoder E 1, a probability distribution P 1 and a generator G 1, wherein the encoder E 1 is a cyclic neural network, track data R 1 in a training set is taken as input, potential representation R R of a track is output, the probability distribution P 1 of a path is a probability distribution model based on Gaussian distribution, the potential representation R R of the track is taken as input, mu= { mu 1,μ2,…,μn } is output to represent the type of the path in a potential space, n is the number of the types, the generator G 1 is a cyclic neural network, the potential representation R R of the track is taken as input, and the reconstructed track is output
Constructing a discriminator D which is a k-layer fully connected neural network, and combining the track data R 1 in the training set and the reconstructed trackAs input, output scoring matrix
S105, inputting the training set into the abnormal track detection model for training, and obtaining the optimized abnormal track detection model.
Step 1, changing the lower bound by track informationAs a loss function, using trajectory data R 1 in the training set, the trajectory potentially representing μ, training the neural network parameters in encoder E 1 and generator G 1 of the user-variational automatic encoder using a random gradient descent method;
Wherein, Θ 1 represents the neural network parameters in the encoder E 1 and generator G 1, respectively, to be trained; representing a trajectory Probability of belonging to normal path;
And 2, cycling the step 1, and training the abnormal track detection model until the loss function is lower than the set threshold value.
S106, inputting the test set into an optimized abnormal track detection model, and obtaining the abnormal score of the track.
Referring to fig. 3, the path probabilities are inferred using the trained model, as follows:
Inputting the trajectory data R 2 in the test set to the encoder section in the optimized abnormal trajectory detection model, the resulting output being a potential representation R R of the trajectory;
the probability of conforming to the path μ k εμ in the potential space is calculated based on the potential representation R R, and the anomaly score s (R) for the trace is calculated from the probability.
S107, judging whether the track is abnormal or not based on the abnormality score.
And judging whether the track is abnormal or not, specifically, comparing s (R) with a specified value, and judging whether the track is abnormal or not, wherein the specified value is a manually set value.
The method for calculating the anomaly score s (R) comprises the following steps:
Where μ c represents the path potential representation in the potential vector space.
Referring to fig. 4, the invention discloses an abnormal track detection system based on a cyclic neural network and a differential self-encoder, which comprises:
The segmentation module is used for segmenting the acquired original track data to acquire segmented track data;
the first dividing module is used for dividing the target map into a plurality of grids, and discretizing the segmented track data to obtain discretized track data;
The second division module is used for dividing the discretized track data to obtain a training set and a testing set;
The construction module is used for constructing an abnormal track detection model based on the cyclic neural network and the differential self-encoder;
the training module is used for inputting the training set into the abnormal track detection model for training to obtain an optimized abnormal track detection model;
the testing module is used for inputting the testing set into the optimized abnormal track detection model to obtain the abnormal score of the track;
and the judging module is used for judging whether the track is abnormal or not based on the abnormal score.
The experimental results of this example are shown in table 1:
ACC F1
SD-Pair1 0.9880 0.9796
SD-Pair2 0.9722 0.9167
SD-Pair3 1.0000 1.0000
SD-Pair4 0.9924 0.9767
SD-Pair5 0.9314 0.8955
the Accuracy (ACC) of the test set stabilized above 0.93 and the F1 value stabilized above 0.89.
The experimental result shows that the abnormal track detection method based on the cyclic neural network and the differential self-encoder can effectively excavate useful features in time information and space information of the running track, further can accurately model abnormal modes of the track, and has high prediction result accuracy, small error and high practical value.
The above is only a preferred embodiment of the present invention, and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (7)

1. The abnormal track detection method based on the cyclic neural network and the differential self-encoder is characterized by comprising the following steps of:
segmenting the acquired original track data to acquire segmented track data, wherein the segmented track data comprises the following specific steps:
acquiring track point position information and time stamp of the vehicle from the driving track data, and based on the track point position information and time stamp of the vehicle, sequencing Represents a track in track data, wherein,Representing sequencesMiddle (f)The time stamps of the individual track points and the position information,Is the length of the track;
Based on track segmentation rule, the original track data Segmentation is performed, using sequencesRepresents a track in track data, wherein,Representing sequencesMiddle (f)The time stamps of the individual track points and the position information,Is the length of the track;
The track segmentation rule specifically comprises the following steps:
When the space deviation of two adjacent track points is larger than a set first threshold value, lagrange interpolation is carried out through surrounding track points to form complete track data, or when the track points do not displace within a set time or the displacement deviation of the track points does not reach a set second threshold value, the track is segmented to form complete track data;
the track is divided, specifically:
Removing the track points which do not displace within the set time, and dividing the displacement points which do not move to obtain segmented track data;
dividing a target map into a plurality of grids, and discretizing the segmented track data to obtain discretized track data;
dividing discretized track data to obtain a training set and a testing set;
Constructing an abnormal track detection model based on the cyclic neural network and the differential self-encoder;
Inputting the training set into an abnormal track detection model for training to obtain an optimized abnormal track detection model;
Inputting the test set into an optimized abnormal track detection model to obtain the abnormal score of the track;
Based on the anomaly score, it is determined whether the trajectory is anomalous.
2. The method for detecting abnormal tracks of the recurrent neural network and the differential self-encoder according to claim 1, wherein the dividing the target map into a plurality of grids, and performing discretization processing on the segmented track data to obtain discretized track data, specifically:
Dividing the target map into a plurality of grids, wherein the track points falling into the same grid have the same number to realize the discretization of the track data points, and using a sequence Represents a track in track data, wherein,Representing sequencesMiddle (f)The time stamps of the individual track points and the position information,Is the length of the track.
3. The abnormal track detection method of the recurrent neural network and the differential self-encoder according to claim 2, wherein the construction of the abnormal track detection model based on the recurrent neural network and the differential self-encoder is specifically as follows:
Step 4.1, constructing a track self-encoder model including an encoder Probability distributionSum generatorEncoder(s)To circulate the neural network, the track data in the training set is used forPotential representations of output trajectories as inputsProbability distribution of pathsFor probability distribution model based on Gaussian distribution, potential representation of traceAs input, outputRepresenting the path category in the potential space, whereinIs the number of the categories, generatorTo cycle the neural network, potential representations of the trajectories are usedAs input, the reconstructed trajectory is output;
Step 4.2 construction of a discriminatorDistinguishing deviceIs thatLayer full-connection neural network, track data in training setAnd reconstructed trajectoryAs input, output scoring matrix。
4. The abnormal track detection method based on the recurrent neural network and the differential self-encoder as claimed in claim 3, wherein the training set is input into the abnormal track detection model for training, and an optimized abnormal track detection model is obtained, specifically:
step 5.1 changing the lower bound with track information Using trajectory data in training sets as a loss functionPotential representation of a trackEncoder for training user variation automatic encoder by adopting random gradient descent methodSum generatorThe neural network parameters of (a); potential representation of trajectories;
Wherein, Respectively represent the encoders to be trainedSum generatorThe neural network parameters of (a); representing a trajectory Probability of belonging to normal path;
And 5.2, training the abnormal track detection model in the step 6.1 in a circulating way until the loss function is lower than a set threshold value.
5. The method for detecting abnormal tracks of the recurrent neural network and the differential self-encoder according to claim 4, wherein the inputting the test set into the optimized abnormal track detection model, obtaining the abnormal score of the track comprises the following specific steps:
Trajectory data in the test set The encoder portion input into the optimized abnormal trajectory detection model, the resulting output being a potential representation of the trajectory;
Based on potential representationComputing paths in conforming potential spaceAnd calculating the anomaly score of the trace from the probabilities。
6. The anomaly trajectory detection method based on a recurrent neural network and differential self-encoder of claim 5, wherein anomaly scoreThe calculation method of (1) is as follows:
Wherein, Representing a path potential representation in a potential vector space.
7. The abnormal track detection system based on the cyclic neural network and the differential self-encoder is characterized by comprising the following components:
the system comprises a segmentation module, a segmentation module and a control module, wherein the segmentation module is used for segmenting acquired original track data to acquire segmented track data, and the function of the segmentation module is realized by the following steps:
acquiring track point position information and time stamp of the vehicle from the driving track data, and based on the track point position information and time stamp of the vehicle, sequencing Represents a track in track data, wherein,Representing sequencesMiddle (f)The time stamps of the individual track points and the position information,Is the length of the track;
Based on track segmentation rule, the original track data Segmentation is performed, using sequencesRepresents a track in track data, wherein,Representing sequencesMiddle (f)The time stamps of the individual track points and the position information,Is the length of the track;
The track segmentation rule specifically comprises the following steps:
When the space deviation of two adjacent track points is larger than a set first threshold value, lagrange interpolation is carried out through surrounding track points to form complete track data, or when the track points do not displace within a set time or the displacement deviation of the track points does not reach a set second threshold value, the track is segmented to form complete track data;
the track is divided, specifically:
Removing the track points which do not displace within the set time, and dividing the displacement points which do not move to obtain segmented track data;
the first dividing module is used for dividing the target map into a plurality of grids, and discretizing the segmented track data to obtain discretized track data;
The second division module is used for dividing the discretized track data to obtain a training set and a testing set;
The construction module is used for constructing an abnormal track detection model based on the cyclic neural network and the differential self-encoder;
the training module is used for inputting the training set into the abnormal track detection model for training to obtain an optimized abnormal track detection model;
the testing module is used for inputting the testing set into the optimized abnormal track detection model to obtain the abnormal score of the track;
and the judging module is used for judging whether the track is abnormal or not based on the abnormal score.
CN202210439506.7A 2022-04-25 2022-04-25 Abnormal trajectory detection method and system based on recurrent neural network and differential autoencoder Active CN114841316B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210439506.7A CN114841316B (en) 2022-04-25 2022-04-25 Abnormal trajectory detection method and system based on recurrent neural network and differential autoencoder

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210439506.7A CN114841316B (en) 2022-04-25 2022-04-25 Abnormal trajectory detection method and system based on recurrent neural network and differential autoencoder

Publications (2)

Publication Number Publication Date
CN114841316A CN114841316A (en) 2022-08-02
CN114841316B true CN114841316B (en) 2025-03-14

Family

ID=82566044

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210439506.7A Active CN114841316B (en) 2022-04-25 2022-04-25 Abnormal trajectory detection method and system based on recurrent neural network and differential autoencoder

Country Status (1)

Country Link
CN (1) CN114841316B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115482446A (en) * 2022-08-30 2022-12-16 电子科技大学 Trajectory Anomaly Detection Method Based on Deep Convolutional Generative Adversarial Network

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111428763A (en) * 2020-03-17 2020-07-17 陕西师范大学 A Pedestrian Trajectory Prediction Method Based on Scene Constraint GAN
CN111694917A (en) * 2020-06-10 2020-09-22 北京嘀嘀无限科技发展有限公司 Vehicle abnormal track detection and model training method and device

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11131993B2 (en) * 2019-05-29 2021-09-28 Argo AI, LLC Methods and systems for trajectory forecasting with recurrent neural networks using inertial behavioral rollout
CN112052873B (en) * 2020-07-30 2022-10-21 北京邮电大学 Abnormal track detection method and device based on universal converter
CN113269115B (en) * 2021-06-04 2024-02-09 北京易航远智科技有限公司 Pedestrian track prediction method based on Informar
CN114090396B (en) * 2022-01-24 2022-06-14 华南理工大学 Cloud environment multi-index unsupervised anomaly detection and root cause analysis method

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111428763A (en) * 2020-03-17 2020-07-17 陕西师范大学 A Pedestrian Trajectory Prediction Method Based on Scene Constraint GAN
CN111694917A (en) * 2020-06-10 2020-09-22 北京嘀嘀无限科技发展有限公司 Vehicle abnormal track detection and model training method and device

Also Published As

Publication number Publication date
CN114841316A (en) 2022-08-02

Similar Documents

Publication Publication Date Title
US11525819B1 (en) Leak detection event aggregation and ranking systems and methods
KR101951595B1 (en) Vehicle trajectory prediction system and method based on modular recurrent neural network architecture
Liang et al. Nettraj: A network-based vehicle trajectory prediction model with directional representation and spatiotemporal attention mechanisms
Rogers et al. Mining GPS data to augment road models
Castignani et al. Smartphone-based adaptive driving maneuver detection: A large-scale evaluation study
Hechifa et al. Improved intelligent methods for power transformer fault diagnosis based on tree ensemble learning and multiple feature vector analysis
CN111209457A (en) Target typical activity pattern deviation warning method
CN118196662B (en) Coal mine digital twin model building method based on Bayesian network algorithm
CN106441279A (en) Robot positioning method based on self-positioning and edge searching and system
CN110895879A (en) Method and device for detecting co-running vehicle, storage medium and electronic device
Bierlaire et al. Modeling route choice behavior from smartphone GPS data
CN118645017A (en) A track prediction method in autonomous operation mode based on time-frequency analysis
CN114841316B (en) Abnormal trajectory detection method and system based on recurrent neural network and differential autoencoder
Ghafouri et al. Optimal detection of faulty traffic sensors used in route planning
Berjisian et al. Evaluation of map‐matching algorithms for smartphone‐based active travel data
Kubota et al. Autopilot model for shield tunneling machines using support vector regression and its application to previously constructed tunnels
CN110716925A (en) Cross-border behavior recognition method based on trajectory analysis
Kattan et al. Sensitivity analysis of an evolutionary-based time-dependent origin/destination estimation framework
Xue et al. Statistical source analysis of recurring sulfur dioxide pollution events in a chemical industrial park
Wu et al. Traffic sensor health monitoring using spatiotemporal graphical modeling
Marciniak et al. Inspection of railway catenary systems using machine learning with domain knowledge integration
Oh et al. Anonymous vehicle tracking for real-time traffic surveillance and performance on signalized arterials
CN114252090B (en) A credibility assessment method for multi-source navigation sensors
Tan et al. Multi-module decision fusion in operational status monitoring
CN118228167A (en) A rail transit running rail grounding fault diagnosis method and system

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