CN113947902A - Real-time traffic accident detection and early warning method based on Seq2Seq self-encoder model - Google Patents
Real-time traffic accident detection and early warning method based on Seq2Seq self-encoder model Download PDFInfo
- Publication number
- CN113947902A CN113947902A CN202111211135.9A CN202111211135A CN113947902A CN 113947902 A CN113947902 A CN 113947902A CN 202111211135 A CN202111211135 A CN 202111211135A CN 113947902 A CN113947902 A CN 113947902A
- Authority
- CN
- China
- Prior art keywords
- traffic
- self
- seq2seq
- window
- 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.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 39
- 206010039203 Road traffic accident Diseases 0.000 title claims abstract description 36
- 238000000034 method Methods 0.000 title claims abstract description 19
- 230000002159 abnormal effect Effects 0.000 claims abstract description 29
- 230000007246 mechanism Effects 0.000 claims abstract description 12
- 238000011897 real-time detection Methods 0.000 claims abstract description 8
- 230000008569 process Effects 0.000 claims abstract description 5
- 230000005856 abnormality Effects 0.000 claims description 7
- 230000002457 bidirectional effect Effects 0.000 claims description 7
- 230000015654 memory Effects 0.000 claims description 7
- 230000002441 reversible effect Effects 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 2
- 238000013507 mapping Methods 0.000 claims description 2
- 230000006403 short-term memory Effects 0.000 claims description 2
- 230000007787 long-term memory Effects 0.000 claims 1
- 238000011160 research Methods 0.000 description 9
- 239000011159 matrix material Substances 0.000 description 6
- 230000003111 delayed effect Effects 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 238000012360 testing method Methods 0.000 description 3
- 238000012549 training Methods 0.000 description 3
- 230000002547 anomalous effect Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000007477 logistic regression Methods 0.000 description 2
- 230000008447 perception Effects 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 238000013058 risk prediction model Methods 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 230000000875 corresponding effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000002955 isolation Methods 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000007634 remodeling Methods 0.000 description 1
- 238000012502 risk assessment Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000013179 statistical model Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0125—Traffic data processing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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
- G06—COMPUTING; CALCULATING OR 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
- G06N3/088—Non-supervised learning, e.g. competitive learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0635—Risk analysis of enterprise or organisation activities
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- General Health & Medical Sciences (AREA)
- Development Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Educational Administration (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Quality & Reliability (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Operations Research (AREA)
- Game Theory and Decision Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Analytical Chemistry (AREA)
- Chemical & Material Sciences (AREA)
- Primary Health Care (AREA)
- Traffic Control Systems (AREA)
Abstract
The application discloses a real-time traffic accident detection and early warning method based on a Seq2Seq self-encoder model. Firstly, in order to analyze traffic element sequence data before and after a traffic accident and the characteristics of traffic flow abnormity, the invention establishes a set of traffic abnormity detection model and early warning process based on a Seq2Seq self-encoder. Secondly, the invention introduces an Attention mechanism on the basis of the Seq2Seq model to capture important traffic state characteristics. Then, in the aspect of judging the abnormal traffic state, the real-time detection of the traffic accident and the classification of accident risk grades are realized by comparing the reconstruction errors of the original data and the predicted data and according to a set threshold value.
Description
Technical Field
The application relates to the technical field of traffic anomaly detection, in particular to a real-time traffic accident detection and early warning method based on a Seq2Seq self-encoder model.
Background
With the rapid development of an Intelligent Transportation System (ITS), a traffic management department can determine real-time data such as traffic flow, speed, traffic density and the like through various vehicle detection equipment, so that the perception degree of the traffic state can be greatly improved, and more comprehensive support is provided for early warning of traffic accidents, accident detection and accident handling.
Chakraborty et al [1] propose a data-driven AID (automatic identification detection) framework that can exploit large-scale historical traffic data and the inherent topology of traffic networks to achieve reliable traffic patterns. Many scholars are engaged in the development of traffic models, mainly for predicting the possibility of accidents with different degrees of severity, and for real-time accident risk prediction by using traffic data collected from detection stations. Kwak and Kho construct a real-time accident risk prediction model aiming at different road section types and traffic flow states of the expressway. In combination with past research, traffic accident risk prediction models can be roughly divided into two categories, namely models based on statistical theory and machine learning models.
Model based on statistical theory: Abdel-Aty et al propose matching case-to-logistic regression models, Xu et al propose binary logit models to demonstrate traffic accident problems, and relate the probability of occurrence of crashes of different severity to various traffic flow characteristics obtained from detector data. Chenyufei and the like research static traffic accident prediction by adopting a grey correlation degree analysis method. Li Rong and the like research a road traffic accident frequency prediction method based on a transcoloric function, and the fitting mode does not need to consider the real-time characteristics of traffic flow. Overall, the past real-time accident risk assessment research mainly adopts a logistic regression statistical model, and such a model has the disadvantage of being in a linear function form.
Content of application
The application aims to provide a real-time traffic accident detection and early warning method based on a Seq2Seq self-encoder model, which utilizes a self-decoder to effectively extract original column characteristics of speed and flow, establishes a set of Seq2Seq self-encoder deep learning model to extract important traffic information and captures or tracks fluctuation or other characteristics under abnormal traffic changes. In addition, the invention respectively determines the prediction and input window errors in each time period, respectively gives the traffic accident threshold value of each road section according to the time period, and finally captures the abnormal traffic state before and after the accident and realizes the early warning and detection of the traffic accident.
In order to achieve the purpose, the application is realized by the following technical scheme:
a real-time traffic accident detection and early warning method based on a Seq2Seq self-encoder model comprises the following steps:
directly extracting original traffic flow and speed sequence data from a Seq2Seq model of an encoder by using an embedded bidirectional long-short term memory network (Bi-LSTM);
in order to meet the network input form of the recurrent neural network in deep learning, a sliding window needs to be constructed and a feature matrix of the traffic flow on a road section needs to be reconstructed. Establishing a data input window in an Autoencoder (Autoencoder), and encoding a feature vector of a flow and speed sequence input in a detection window W at a certain moment into a semantic vector C with a fixed size through a bidirectional long-short term memory network (Bi-LSTM) Autoencoder so as to guide the generation of a predicted value of each step in a prediction sequence;
taking the final hidden layer state H of the nonlinear mapping of the long short-term memory network (LSTM) in the reverse direction of the encoder as the initial state of the decoder;
the bidirectional long-short term memory network (Bi-LSTM) self-decoder continuously generates dynamic semantic vector C through the adjustment of an intermediate Attention Mechanism (Attention Mechanism)i；
The decoder uses the decoder initial state, semantic vector CiAnd step length siEach time point recurrently generates a continuous characteristic sequence in the original window, thereby obtaining a predicted reconstruction window W0；
Calculating a detection window W and a prediction window W at each moment0The total mean square error MSE of the reconstruction error is taken as the reconstruction error;
and setting a threshold value, and judging whether the window belongs to an abnormal window or a normal window. The features under normal and abnormal traffic states are significantly different, so the weight distribution of the normal traffic state feature and the abnormal traffic state feature is different, that is, the weight distribution of the normal traffic state feature extracted by the attention mechanism is a representative feature, and the weight distribution of the abnormal traffic state feature is a non-representative feature. Therefore, the reconstruction error MSE of the model is very large for the input abnormal traffic sequence data. Beyond the decision threshold, the model will identify the time window as an anomalous window. This indicates that there may be an abnormality in the traffic state, and thus the purpose of detecting the abnormal traffic state is more effectively achieved.
Optionally, the traffic element sequence data before and after the traffic accident and the characteristics of traffic flow abnormity are analyzed, and then a set of traffic abnormity detection model and early warning process based on the Seq2Seq self-encoder are established. Firstly, in the aspect of traffic state perception, an established Seq2Seq self-coding model and an Attention mechanism are introduced, so that the capture of important traffic state characteristics is realized; secondly, in the aspect of traffic state abnormity judgment, sequence data input by a Seq2Seq self-encoder reconstruction is utilized, a structure reconstruction error can be obtained by comparing original data, then judgment of a traffic early warning grade and real-time detection of a traffic accident are realized according to a set threshold value, and finally early warning and detection efficiency of traffic accident abnormity detection are improved.
Optionally, an Attention mechanism is introduced from a decoder, and in consideration of continuous abnormity of the traffic state, an intermediate semantic vector with an indefinite length is output to capture important traffic state characteristics and sequence information, in particular to capture traffic characteristics extraction of two traffic elements, namely traffic flow and speed, as two-dimensional characteristic input.
Optionally, in the aspect of judging the traffic state abnormality, the real-time detection of the traffic accident and the classification of the accident risk level are realized by comparing the reconstruction errors of the original data and the predicted data and according to the threshold value.
Optionally, the sequence data is reversely input in the input window to increase the difficulty of reconstruction of abnormal data and significantly strengthen the abnormal traffic wave, thereby improving the capacity of detecting abnormal traffic.
Compared with the prior art, the method has the following advantages:
the invention utilizes the original column characteristics of the speed and the flow effectively extracted by the self-decoder to establish a set of Seq2Seq self-encoder deep learning model to extract important traffic information and capture or track the fluctuation or other characteristics under the abnormal traffic change. In addition, the invention respectively determines the prediction and input window errors in each time period, respectively gives the traffic accident threshold value of each road section according to the time period, and finally captures the abnormal traffic state before and after the accident and realizes the early warning and detection of the traffic accident.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings without creative efforts.
FIG. 1 is a traffic accident early warning design flow in the embodiment of the present application;
FIG. 2 is a data input-model fitting-reconstruction process of a Seq2Seq self-encoder in an embodiment of the present application;
fig. 3 is a traffic anomaly identification model based on Seq2Seq in the embodiment of the present application.
Detailed Description
In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described with reference to the accompanying drawings, the described embodiments should not be considered as limiting the present application, and all other embodiments obtained by a person of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.
In the following description, reference is made to "some embodiments," "one or more embodiments," which describe a subset of all possible embodiments, but it is understood that "some embodiments," "one or more embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict.
In the following description, references to the terms "first \ second \ third" are used for respective similar objects only and do not denote a particular order or importance to the objects, it being understood that "first \ second \ third" may be interchanged under certain circumstances or sequences of events to enable embodiments of the application described herein to be practiced in other than those illustrated or described.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of the present application only and is not intended to be limiting of the application.
The embodiment provides a real-time traffic accident detection and early warning method based on a Seq2Seq self-encoder model. Wherein, the Seq2Seq is formed by combining an encoding layer, a semantic vector and a decoding layer. Since the input and output of the network are the same, in order to improve the ability of the network to reshape the window and the fitting effect, the output can be set as the reverse sequence of the detection window. Since the sequence generated by the Seq2Seq decoder at each moment in decoding is highly correlated with the previous moment, it is difficult to generate the forward direction for the model to generate the reverse input window. In this case, if a traffic anomaly window is input, the Seq2Seq self-decoder will have difficulty in identifying the anomalous data characteristics and cause the sequence error generated at each time to become larger and larger. In addition, the study introduces a bahdana attention mechanism to enhance the memory of the network and better acquire important information of long sequences. The data input-model fitting-reconstruction process of Seq2Seq self-encoder is shown in fig. 2.
(1) Network input feature matrix
The model is firstly used for early warning of potential accidents of the road section, a sliding window is constructed and a characteristic matrix of the traffic flow on the road section is reconstructed in order to meet the network input form of the recurrent neural network in deep learning: remodeling the original data form into 3 dimensions (section flow, lane flow and driving speed), defining the data of the previous 20 minutes of each time point as a window characteristic matrix of the time, wherein L is the length of each window, the data sets are different, and the setting of model parameters is different. In order to find the optimal accident detection effect, the difference of the generalization capability of the model is obtained by observing the detection windows with different traffic flow sequence lengths L input by the model in the training stage. The step length s is set to 1 to ensure the time continuity of the window at each moment and is constructed in sequence in the form of a sliding window.
(2) Index abnormality recognition model
The invention applies the Seq2Seq self-coding network to the detection of traffic state abnormity, and the essential basis is whether the difference size of the original window and the reconstructed window meets the threshold requirement. In the encoder portion, successive traffic flow sequences are encoded in sequence:
the input sequence is as follows: x ═ X1,x2,x3,x4···,xt-1,xt}，
The output sequence of the decoder is:
(3) threshold determination
Taking the total Mean square error MSE (Mean squared error, MSE) as the reconstruction error:
wherein: t ═ T1-T0Is a prediction time interval;
is a predicted value at the time t; xtIs the true value at time t. The larger the reconstruction error MSE of the window is, the lower the structural similarity between the original detection window and the remolded window is, the more the structural similarity exceeds the judgment threshold value, the model identifies the window at the moment as an abnormal window, and the traffic state is possibly abnormal, namely the accident risk is increased. The features under normal and abnormal traffic states are obviously different, so the weight distribution of the normal traffic state feature weight distribution and the abnormal traffic state feature weight distribution are different, namely, the weight distribution of the normal traffic state feature weight extracted by the attention mechanism is a representative feature, and the weight distribution of the abnormal traffic state feature weight is a non-representative featureAnd (4) characteristic features.
A large number of documents are the roads of highways with strong closure (physical isolation), other roads of lower grade do not directly cause anomalies in their traffic. Although the elevated road has strong closure, the up-and-down ramps may induce the problem of continuous interleaving or traffic overflow during the morning and evening peak hours, and may directly affect the traffic state of the elevated main road. In addition, the operation speed of the elevated road is lower than that of the highway in terms of the significance of the accident, and the severity of the accident and the traffic influence are small. Traffic accidents on highways are relatively serious, the accident handling time is long, and the traffic influence is large, such as the vehicle running speed changes more obviously. Therefore, the accident characteristics of the elevated road are less significant than those of the expressway. Therefore, the data noise of the elevated road is relatively larger, and the detection difficulty is also larger.
Existing research has been focused on anomaly detection using a symmetric autoencoder network structure, and the field of machine translation is strictly adopted in this document as a Seq2Seq network: the encoder is different from other researches, and the Bi-LSTM is adopted to more effectively extract the context time sequence characteristics of the normal traffic state; the reconstruction of the original data in the decoding stage is to generate sequences one by one according to time steps, so if abnormal data is input in a test, the accumulated sequence reconstruction error is very large. Secondly, the existing automatic encoder research rarely uses the attention mechanism, or the attention mechanism is poor in effect and interpretability, and the attention mechanism is applied to capturing of normal/abnormal traffic states and is deeply analyzed.
(4) Demonstration of experiments
The Shanghai Yanan elevated road is a trunk elevated expressway in an east-west trend, the total length is 13 kilometers, three sections of the trunk elevated expressway are built to cross urban areas and mainly comprise six bidirectional lanes, and the designed speed per hour is 80 km/h. 5 upper ramps and 8 lower ramps are arranged in the east-west direction; the west is provided with 6 upper ramps and 7 lower ramps upward. The experimental data demonstrated this time are as follows:
data one: delay and safety of the traffic accident data of the overhead release section;
the data mainly comprises 94 pieces of traffic accident data which occur within two weeks of the delayed and elevated height, wherein the data comprises the starting time of the accident, the ending time of the accident, a road section code where the accident is located, a position description, an elevated road name and an accident description.
Data II: and delaying the flow speed data of the overhead issuing section.
The data of the flow speed of the distribution section of the delayed safety overhead comprises 401906 pieces of data of two weeks of 38 road sections of the delayed safety overhead and 2 tunnel sections of the delayed safety east road.
Considering the difference in the characteristics of the data, the present study divides the above data into a normal traffic state data set and an abnormal traffic state set. In addition, the normal section length (ramp section), the coil distribution and the traffic space-time characteristics of the delay elevated road are different, the whole data can be divided into 8 data subsets, and the corresponding training set and the traffic accident judgment threshold value are determined.
The Seq2Seq automatic coding and decoding network carries out real-time early warning and prejudgment on possible accidents of all sections of the delayed viaduct at each moment or in a short time later. In order to verify the effectiveness of the anomaly detection model, sample data of 1 hour before and after all two/multiple vehicle accidents occur within 2 weeks of the original data are brought into the trained model for testing. In addition, with the aid of the statistical idea of the 3 σ criterion, thresholds are defined for the training phase, which use the quantiles of 90%, 95%, and 99% of the window error exceeding the respective time instants as the test phase, respectively, as shown in table 1.
TABLE 1 threshold level division for different time periods in the south and north sides of Yanan elevated
Accident detection is the identification of abnormal traffic conditions for the duration of a traffic accident. The traditional binary classification model generally carries out 0-1 label classification prediction on normal or abnormal accident data directly. The accident detection in the research is the popularization of a Seq2Seq self-encoding and decoding model, and the structural errors of the original detection window and the prediction window also need to be compared. When the total window error at a certain moment is greater than the early warning level threshold value calibrated in the time interval, the prediction label at the moment is 1 (abnormal), otherwise, the prediction label is 0 (normal).
In order to verify the accident detection effect of the model and measure the generalization ability of the model, the research adopts a standard confusion matrix in the classification indexes as an evaluation mode. The confusion matrix can obtain that the recall rate of the accident sample data on the north side of the Yanan elevated is 72.7%, the false alarm rate (false positive rate FPR) is 21.1%, the recall rate of the accident sample data on the south side is 76.4%, and the false alarm rate is 22.8%.
Compared with the prior art, the invention can demonstrate higher sensitivity and lower false alarm rate on the detection of the traffic accident, and the specific results are shown in table 2.
TABLE 2 sensitivity and false alarm Rate comparison for real-time Accident detection
The above description is only an example of the present application, and is not intended to limit the scope of the present application. Any modification, equivalent replacement, and improvement made within the spirit and scope of the present application are included in the protection scope of the present application.
Claims (5)
1. A real-time traffic accident detection and early warning method based on a Seq2Seq self-encoder model is characterized by comprising the following steps:
directly extracting original traffic flow and speed sequence data by using a Seq2Seq model embedded into a bidirectional long and short term memory network self-encoder;
establishing a data input window in a self-encoder, and encoding the characteristic vector of the flow and speed sequence input in a detection window W at a certain moment into a semantic vector C with a fixed size through a bidirectional long-short term memory network self-encoder so as to guide the output of the predicted value of each step in a prediction sequence;
taking the final hidden layer state H of the nonlinear mapping of the long-short term memory network reversed by the encoder as the initial state of the decoder;
the bidirectional long-short term memory network self-decoder continuously generates dynamic semantic vector C through the adjustment of an intermediate attention mechanismi；
The decoder uses the decoder initial state, semantic vector CiAnd step length siEach time point recurrently generates a continuous characteristic sequence in the original window, thereby obtaining a predicted reconstruction window W0；
Calculating a detection window W and a prediction window W at each moment0The total mean square error MSE of the reconstruction error is taken as the reconstruction error; and setting a threshold value, and judging whether the window belongs to an abnormal window or a normal window.
2. The method for real-time detection and early warning of traffic accidents based on the Seq2Seq self-encoder model as claimed in claim 1, wherein the traffic element sequence data before and after the traffic accidents and the characteristics of traffic flow abnormality are analyzed first, and then a set of traffic abnormality detection model and early warning process based on the Seq2Seq self-encoder is established.
3. The real-time detection and early warning method for traffic accidents based on the Seq2Seq self-encoder model according to claim 1, characterized in that an Attention mechanism is introduced into a self-decoder, and in consideration of the continuous abnormality of the traffic state, an intermediate semantic vector with an indefinite length is output to capture important traffic state features and sequence information, especially to capture traffic characteristics extraction of two-dimensional feature input of two traffic elements of traffic flow and speed.
4. The method for real-time detection and early warning of traffic accidents based on the Seq2Seq self-encoder model as claimed in claim 1, wherein the real-time detection of traffic accidents and classification of accident risk levels are achieved by comparing the reconstruction errors of the original data and the predicted data and according to the threshold in the aspect of the abnormal traffic status determination.
5. The real-time detection and early warning method for traffic accidents based on the Seq2Seq self-encoder model as claimed in claim 1, wherein the sequence data is inputted in a reverse direction in the input window to increase the difficulty of reconstruction of abnormal data and to significantly reinforce abnormal traffic waves, thereby improving the ability of traffic abnormality detection.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111211135.9A CN113947902A (en) | 2021-10-18 | 2021-10-18 | Real-time traffic accident detection and early warning method based on Seq2Seq self-encoder model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111211135.9A CN113947902A (en) | 2021-10-18 | 2021-10-18 | Real-time traffic accident detection and early warning method based on Seq2Seq self-encoder model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113947902A true CN113947902A (en) | 2022-01-18 |
Family
ID=79331298
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111211135.9A Pending CN113947902A (en) | 2021-10-18 | 2021-10-18 | Real-time traffic accident detection and early warning method based on Seq2Seq self-encoder model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113947902A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114004429A (en) * | 2022-01-04 | 2022-02-01 | 苏州元澄科技股份有限公司 | Data processing method and system for constructing digital city |
| CN115018213A (en) * | 2022-08-09 | 2022-09-06 | 深圳市城市交通规划设计研究中心股份有限公司 | Expressway accident duration prediction method, electronic device, and storage medium |
| CN115964636A (en) * | 2022-12-23 | 2023-04-14 | 浙江苍南仪表集团股份有限公司 | Gas flow abnormity detection method and system based on machine learning and dynamic threshold |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2023308A1 (en) * | 2007-07-25 | 2009-02-11 | Hitachi Ltd. | Traffic incident detection system |
| CN109993970A (en) * | 2019-03-15 | 2019-07-09 | 西南交通大学 | A kind of city area-traffic accident risk prediction method |
| CN112784121A (en) * | 2021-01-28 | 2021-05-11 | 浙江工业大学 | Traffic accident prediction method based on space-time diagram representation learning |
| CN113222218A (en) * | 2021-04-16 | 2021-08-06 | 浙江工业大学 | Traffic accident risk prediction method based on convolution long-time and short-time memory neural network |
-
2021
- 2021-10-18 CN CN202111211135.9A patent/CN113947902A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2023308A1 (en) * | 2007-07-25 | 2009-02-11 | Hitachi Ltd. | Traffic incident detection system |
| CN109993970A (en) * | 2019-03-15 | 2019-07-09 | 西南交通大学 | A kind of city area-traffic accident risk prediction method |
| CN112784121A (en) * | 2021-01-28 | 2021-05-11 | 浙江工业大学 | Traffic accident prediction method based on space-time diagram representation learning |
| CN113222218A (en) * | 2021-04-16 | 2021-08-06 | 浙江工业大学 | Traffic accident risk prediction method based on convolution long-time and short-time memory neural network |
Non-Patent Citations (1)
| Title |
|---|
| 赵超等: "基于Seq2Seq 自编码器模型的交通事故实时检测与评价", 《控制与决策》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114004429A (en) * | 2022-01-04 | 2022-02-01 | 苏州元澄科技股份有限公司 | Data processing method and system for constructing digital city |
| CN115018213A (en) * | 2022-08-09 | 2022-09-06 | 深圳市城市交通规划设计研究中心股份有限公司 | Expressway accident duration prediction method, electronic device, and storage medium |
| CN115964636A (en) * | 2022-12-23 | 2023-04-14 | 浙江苍南仪表集团股份有限公司 | Gas flow abnormity detection method and system based on machine learning and dynamic threshold |
| CN115964636B (en) * | 2022-12-23 | 2023-11-07 | 浙江苍南仪表集团股份有限公司 | Gas flow abnormality detection method and system based on machine learning and dynamic threshold |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113947902A (en) | Real-time traffic accident detection and early warning method based on Seq2Seq self-encoder model | |
| CN107591002B (en) | Real-time estimation method for highway traffic parameters based on distributed optical fiber | |
| CN106874863B (en) | Vehicle illegal parking and reverse running detection method based on deep convolutional neural network | |
| Dong et al. | Improved robust vehicle detection and identification based on single magnetic sensor | |
| CN102819952B (en) | Method for detecting illegal lane change of vehicle based on video detection technique | |
| CN106845424B (en) | Pavement remnant detection method based on deep convolutional network | |
| CN104200669B (en) | Fake-licensed car recognition method and system based on Hadoop | |
| Coifman | Vehicle re-identification and travel time measurement in real-time on freeways using existing loop detector infrastructure | |
| CN105809679B (en) | Mountain railway side slope rockfall detection method based on visual analysis | |
| CN102682601B (en) | Expressway traffic incident detection method based on optimized support vector machine (SVM) | |
| CN103456175A (en) | Accompanying vehicle real-time detection method based on vehicle registration plate recognition and meshing monitoring | |
| CN114049765B (en) | Urban road network traffic flow OD estimation method based on automatic vehicle number plate identification data | |
| CN110111578B (en) | Tensor recovery-based sporadic traffic congestion detection method | |
| CN102663833A (en) | Method for identifying traffic events for special gate of urban railway transit | |
| CN111768619A (en) | Express way vehicle OD point determining method based on checkpoint data | |
| CN111667204A (en) | Method and system for determining and grading environmental risk degree of automatic driving open test road | |
| CN112541480B (en) | Online identification method and system for tunnel foreign matter invasion event | |
| CN104156691A (en) | Monitoring method based on picture processing for detecting behavior of pedestrian climbing over turnstile | |
| CN117609885A (en) | High-speed rail intrusion monitoring method based on distributed optical fiber sensing and fused neural network | |
| CN101482981A (en) | License event recognition method for special shearing type door gate machine used in urban track traffic | |
| CN103208010B (en) | A kind of traffic state quantitative identification method of view-based access control model feature | |
| CN118015562A (en) | Method and system for extracting key frames of traffic accident monitoring video in severe weather | |
| CN104392076A (en) | Urban road network pedestrian traffic accident black spot identification method | |
| CN112183597B (en) | Unbalanced traffic accident severity analysis method based on spectral clustering | |
| Zhang et al. | Robust vehicle detection and identification with single magnetic sensor |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20220118 |
|
| RJ01 | Rejection of invention patent application after publication |