WO2022241802A1

WO2022241802A1 - Short-term traffic flow prediction method under complex road network, storage medium, and system

Info

Publication number: WO2022241802A1
Application number: PCT/CN2021/095955
Authority: WO
Inventors: 黄社阳; 陈建良; 黄宇恒; 金晓峰; 杨振
Original assignee: 广州广电运通金融电子股份有限公司
Priority date: 2021-05-19
Filing date: 2021-05-26
Publication date: 2022-11-24
Also published as: CN113240182A

Abstract

Provided are a short-term traffic flow prediction method under a complex road network, a storage medium, and a system. The method comprises: collecting traffic flow data of a plurality of road segments; performing data cleaning and time slicing on the traffic flow of the plurality of road segments to obtain multi-dimensional traffic flow data; on the basis of short-term traffic flow model training, training the traffic flow data using a long short-term memory network; and on the basis of a trained short-term traffic flow model, performing short-term traffic flow prediction on the plurality of road segments under the complex road network using the long short-term memory network. The method, under the complex road network, can directly perform, by means of a long short-term memory network classifier, short-term traffic flow prediction on N road segments to be predicted, and is not limited to traffic flow prediction of a single upstream or downstream road segment, and the processing method is simple and effective; the method has strong learning and memory capacities and a good prediction effect for short-term traffic flow training and prediction, and has strong algorithm robustness, real-time performance, and strong applicability.

Description

Short-term traffic flow prediction method, storage medium and system under complex road network

technical field

The invention belongs to the field of smart cities and smart security, relates to smart traffic technology, and specifically relates to a short-term traffic flow prediction method, storage medium and system under complex road networks.

Background technique

The Intelligent Transportation System (ITS for short) is a real-time, accurate and efficient comprehensive transportation management system that functions in a wide range and in all directions and is effectively integrated and applied to the entire ground traffic management system. Constructing the traffic flow guidance subsystem in ITS is one of the most effective ways to solve urban traffic congestion and improve the efficiency of road network traffic. To achieve real-time traffic control and guidance in ITS, it is necessary to have timely and accurate short-term traffic information. Flow prediction provides support for it, so short-term traffic flow prediction has become a research hotspot in intelligent transportation systems. Short-term traffic flow prediction is a kind of prediction in a microscopic sense, and it is one of the core contents of intelligent transportation systems. It has the characteristics of high nonlinearity and uncertainty, and is correlated with time and space. The research shows that the traffic volume at a certain moment on a traffic section in the urban AC road network is related to the traffic flow of the previous sections of this section, and the traffic flow has quasi-periodic characteristics. related to traffic flow.

The commonly used short-term traffic flow forecasting methods include hidden Markov forecasting, BP neural network model forecasting, CNN neural network forecasting, etc.

Short-term traffic flow forecasting method based on hidden Markov:

This method uses the traffic flow information of the predicted road section, based on the current traffic flow information and historical traffic flow information, and uses the hidden Markov algorithm to predict the traffic flow information of the area in the next time period.

Short-term traffic flow forecasting method based on BP neural network:

This method uses the traffic flow information of the predicted road section, based on the current traffic flow information and historical traffic flow information, and uses the BP neural network algorithm to predict the traffic flow information of the area in the next time period.

Short-term traffic flow forecasting method based on convolutional neural network:

This method uses the historical traffic data of the predicted road section and its upstream and downstream sections in the data set to form an input matrix. The flow data of the upstream section is used as the upper half of the input matrix, and the flow data of the downstream section is used as the lower half of the input matrix. The traffic data is placed in the middle, and the convolutional neural network algorithm is used to predict the traffic flow.

The factors that affect the traffic flow forecasting results mainly include time factor and space factor. The existing traffic flow forecasting methods consider more time factors than space factors, which affects the accuracy of traffic flow forecasting.

Secondly, under the complex urban road network, only considering the short-term traffic flow of one road section and predicting the traffic flow of the next time period, there is a problem that the information utilization is incomplete and affects the prediction accuracy. The existing short-term traffic flow forecasting methods mainly include traditional neural network forecasting algorithms such as BP, and the deep learning algorithm is rarely introduced, and the algorithm has limited ability to express data.

Contents of the invention

In order to overcome the deficiencies of the prior art, the present invention provides a short-term traffic flow prediction method, storage medium and system under complex road networks, which can solve the problem of insufficient consideration of spatial internal factors in existing traffic flow prediction methods .

Design principle: Because the short-term traffic flow prediction is affected by two aspects: time factor and space factor. Therefore, this scheme proposes to use the historical flow data of N predicted road sections, and use the long-term short-term memory network to predict the traffic flow data.

Design scheme: in order to solve the aforementioned problems, the specific scheme of the present invention is as follows.

A short-term traffic flow prediction method under a complex road network, the method comprising:

S1. Multi-section traffic flow data collection;

S2. Perform data cleaning and time slicing on the multi-section traffic flow obtained in step S1 to obtain multi-dimensional traffic flow data;

S3. Short-term traffic flow model training, using long-term short-term memory network to train traffic flow data;

S4. Prediction of short-term traffic flow model. Based on the trained short-term traffic flow model, long-short-term memory network is used to predict short-term traffic flow of multiple road sections under complex road network.

Further, step S1 includes: S11, selecting N road sections, where N is a positive integer ≥ 2, and the road sections include intersections and/or road section points; S12, triggering the selected road sections through ground induction coils, video triggers or GPS The data collects the traffic flow data of the N road sections.

Further, step S2 includes: S21, performing data cleaning on the traffic flow data collected in step 1; S22, time slicing the data after data cleaning at a set time interval t, and cutting the data of each road section into M time intervals segment to generate N*M traffic flow data matrix.

Further, step S3 includes:

S31. Traffic flow data preprocessing, normalize the traffic flow data obtained in step S2, the normalized traffic flow x _i ' at the current moment is:

In the formula, x _i ′ is the normalized traffic flow at the current moment, x _max is the maximum value of the traffic flow at the historical moment, and x _min is the minimum value of the traffic flow at the historical moment;

S32. Input the normalized N*M traffic flow data matrix as a training sample into the long-short-term memory network model for training.

Further, step S32 includes:

S321, the matrix extracts advanced features through the convolution layer;

S322, pooling layer processing, to reduce the number of parameters and control overfitting;

S323, activation function processing, using ReLU activation function processing to avoid mode collapse;

S324, Dropout layer processing, to alleviate the occurrence of overfitting;

S325. Based on the attention mechanism, learn the associated information of the time series data through the LSTM layer of the long short-term memory network to improve the model training effect;

S326. Evaluate the prediction effect of the model training on the data through a loss function.

Further, the attention-based long-short-term memory network LSTM layer processing of step S325 includes:

The output o of the long-term short-term memory network LSTM is weighted and summed with the layer weight W, and the attention coefficient a is obtained through softmax, and a is mapped to the output value through the fully connected layer:

a＝softmax(w ^T sigmoid(o))……………………………Formula 2

output＝sigmoid(w ^T (oa ^T ))……………………………Formula 3

In the formula, o is the output of LSTM, and W is the weight of the first layer.

Further, the loss function in step S326 adopts the mean square error loss, and this statistical parameter is the mean value of the sum of squares of the corresponding point errors between the predicted data and the original data, and the calculation formula is:

In the formula, y _i represents the true value of the i-th sample;

f( _xi ) represents the predicted value of the i-th sample;

n is the number of samples.

Further, the short-term traffic flow model prediction in step S4 includes:

S41. Collect the traffic flow data of the previous M time periods, and perform data normalization processing;

S42. Input the normalized traffic flow data matrix into the long-short-term memory network model to predict the short-term traffic flow of multiple road sections.

The present invention also provides a computer-readable storage medium, on which computer-executable program instructions are stored, and the instructions are executed by a processor and a memory, so as to realize the foregoing method.

The present invention also provides a short-term traffic flow prediction system under a complex road network, the system includes a data acquisition module connected by telecommunication, a data preprocessing module, a short-term traffic flow training prediction module, a prediction effect evaluation module and a traffic flow display module ;

The data acquisition module includes a data acquisition unit and a timer; the data acquisition unit includes a ground sensing coil, a video monitor or a GPS assembly, and under the control of the timer, the acquisition unit synchronously collects a plurality of road sections at equal time intervals traffic flow data;

The data preprocessing module receives the traffic flow data of the data acquisition module, and performs data cleaning and time slicing to obtain a traffic flow data matrix;

The short-term traffic flow training prediction module includes a long-short-term memory network training unit and a prediction unit based on the attention mechanism, the training unit is used to train the historical traffic flow data matrix, and the prediction unit receives the current traffic flow data matrix Predict the traffic flow data in the next period;

The prediction effect evaluation module evaluates the accuracy of the short-term traffic flow training prediction module through a loss function;

The traffic flow display module visually displays historical data, current real data, forecast data and forecast effects of multiple road sections.

Compared with the prior art, the beneficial effects of the present invention are:

1. Proposed to add the traffic flow data of N sections of the intersecting road network, including the data of multiple traffic intersections, fully considering that the traffic flow is not only related to the change of flow value on the upstream and downstream roads, but also the traffic flow data of other roads connected to it , it better solves the problem of less consideration of spatial factors in short-term traffic flow prediction under urban complex road network, and has a positive effect on traffic flow prediction.

2. The long-short-term memory network algorithm based on the attention mechanism is used to conduct model training and prediction on the collected traffic flow data, which improves the expression ability of the prediction model and the effect of model prediction.

Description of drawings

Fig. 1 is the overall flowchart of the short-term traffic flow prediction method under the complex road network of the present invention;

Figure 2 is a short-term traffic flow data collection diagram under the complex urban road network;

Fig. 3 is a flow chart of long short-term memory network model training;

Fig. 4 is the prediction flow chart of long short-term memory network model based on attention mechanism;

Fig. 5 is a schematic diagram of a short-term traffic flow prediction system under a complex road network.

Detailed ways

In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments It is a part of embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

It should be understood that "system", "device", "unit" and/or "module" used in this specification is a method for distinguishing different components, elements, parts, parts or assemblies of different levels. However, the words may be replaced by other expressions if other words can achieve the same purpose.

first embodiment

A short-term traffic flow prediction method under a complex road network, see Fig. 1, the method includes the following steps.

S1. Multi-section traffic flow data collection. Step S1 includes:

S11. Select N road sections, where N is a positive integer ≥ 2, and the road sections include intersections and/or road section points;

S12. Collect the traffic flow data of the N road sections for the selected road sections through ground induction coil triggering, video triggering or GPS data.

The methods of traffic flow data collection include: ground sense coil trigger collection, video trigger collection and GPS data collection and other methods. Due to the complex urban road network, wide monitoring scenarios, and various subjects involved in traffic, it is difficult to use traditional methods to accurately collect traffic flow. Considering that the infrastructure for real-time video monitoring of current traffic conditions is relatively mature, this method uses video trigger The collection method collects and monitors the traffic flow data in the road network.

S2. Perform data cleaning and time slicing on the multi-section traffic flow obtained in step S1 to obtain multi-dimensional traffic flow data. Referring to Fig. 2, step S2 includes:

S21. Perform data cleaning on the traffic flow data collected in step 1. Among them, data cleaning includes: analyzing the number of vehicles in the data, time information, whether the collection points are valid, etc., and removing invalid data in the data set.

S22. Perform time slicing on the cleaned data at a set time interval t, and cut the data of each road segment into M time segments, so as to generate an N*M traffic flow data matrix.

Specifically, data collection is performed on the monitored N road sections at an interval of 5 seconds; finally, data preprocessing is performed to obtain N-dimensional traffic flow data.

S3, short-term traffic flow model training, using a long-term short-term memory network to train the traffic flow data; step S3 includes:

S32. Input the normalized N*M traffic flow data matrix as a training sample into the long-term short-term memory network model for training, specifically using the traffic flow data of the previous M moments and N road sections to obtain the N*M data matrix as Training sample input; advanced features are extracted through the convolutional layer, the pooling layer reduces the number of parameters and controls overfitting, the ReLU activation function avoids mode collapse, Dropout alleviates the occurrence of overfitting, and the long-term short-term memory network (LSTM) is better Learn the associated information of time series data, and evaluate the prediction effect of model training through a mean square loss function. Referring to Fig. 3, the training steps of step S32 include:

S321, the matrix extracts advanced features through the convolution layer;

S324, Dropout layer processing, to alleviate the occurrence of overfitting;

S325. Based on the attention mechanism, learn the association information of the time series data through the LSTM layer of the long short-term memory network, so as to improve the model training effect. This is because in traffic flow prediction, when long sequences are considered, the dependencies will be bottlenecked, so the model after the introduction of attention mechanism is required. Specifically, the attention-based long-short-term memory network LSTM layer processing of step S325 includes:

a＝softmax(w ^T sigmoid(o))……………………………Formula 2

output＝sigmoid(w ^T (oa ^T ))……………………………Formula 3

S326. Evaluate the prediction effect of the model training on the data through a loss function. The loss function of step S326 adopts the mean square error loss, and this statistical parameter is the mean value of the sum of squares of the corresponding point errors of the predicted data and the original data, and the calculation formula is:

In the formula, y _i represents the true value of the i-th sample;

f( _xi ) represents the predicted value of the i-th sample;

n is the number of samples.

S4. Prediction of short-term traffic flow model. Based on the trained short-term traffic flow model, long-short-term memory network is used to predict short-term traffic flow of multiple road sections under complex road network. The trained model is used as a prediction model, see Fig. 4, the short-term traffic flow model prediction of step S4 includes:

This method can directly predict the short-term traffic flow of N predicted road sections through the long-short-term memory network classifier in the case of a complex road network, and is not limited to the traffic flow prediction of a single upstream and downstream road section. The processing method is simple and effective. The method proposed in this program has strong learning, memory and prediction effects for short-term traffic flow training and prediction of predicted road sections under complex road networks. The short-term traffic flow prediction method under the complex road network proposed in the scheme has strong algorithm robustness, can achieve real-time performance, and has strong practicability.

second embodiment

The present invention also provides a computer-readable storage medium, on which computer instructions are stored, and the steps of the aforementioned method are executed when the computer instructions are run. Wherein, for the method, please refer to the detailed introduction in the foregoing part, and details will not be repeated here.

Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above-mentioned embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the computer-readable medium includes permanent Both non-permanent and non-permanent, removable and non-removable media can be implemented by any method or technology for information storage. Information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory (ROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Flash memory or other memory technology, Compact Disc Read-Only Memory (CD-ROM), Digital Versatile Disc (DVD) or other optical storage, Magnetic tape cartridge, tape magnetic disk storage or other magnetic storage device or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media excludes transitory computer-readable media, such as modulated data signals and carrier waves.

third embodiment

The present invention also provides a short-term traffic flow prediction system under a complex road network, as shown in Figure 5, the system includes a telecommunications-connected data acquisition module 1, a data preprocessing module 2, a short-term traffic flow training prediction module 3, and prediction effects An evaluation module 4 and a traffic flow display module 5 . in:

The data collection module includes a data collection unit and a timer. The data acquisition unit includes a ground induction coil, a video monitor or a GPS component, and under the control of a timer, the acquisition unit collects traffic flow data of multiple road sections synchronously at equal intervals.

The data preprocessing module receives the traffic flow data from the data acquisition module, and performs data cleaning and time slicing to obtain a traffic flow data matrix.

The short-term traffic flow training prediction module includes a long-short-term memory network training unit and a prediction unit based on the attention mechanism, the training unit is used to train the historical traffic flow data matrix, and the prediction unit receives the current traffic flow data matrix Predict traffic flow data for the next period.

The prediction effect evaluation module evaluates the accuracy of the short-term traffic flow training prediction module through a loss function.

In summary, the design points of the present invention are:

1. Propose N-dimensional traffic flow data collection for N predicted road sections;

2. Proposed short-term traffic flow prediction based on long short-term memory network;

3. Propose short-term traffic flow prediction using long short-term memory network based on attention mechanism;

4. The method of collecting N-dimensional traffic flow data of N predicted road sections is combined with the short-term traffic flow prediction method based on the long-term short-term memory network algorithm based on the attention mechanism, making full use of the long-term short-term memory network algorithm in time and space The powerful learning ability of data improves the forecasting effect, is simple and effective, and has strong practicability.

It should be noted that different embodiments may have different beneficial effects. In different embodiments, the possible beneficial effects may be any one or a combination of the above, or any other possible beneficial effects.

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: it can still be Modifications are made to the technical solutions described in the foregoing embodiments, or equivalent replacements are made to some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

A short-term traffic flow prediction method under a complex road network, characterized in that the method includes:

S1. Multi-section traffic flow data collection;

S2. Perform data cleaning and time slicing on the multi-section traffic flow obtained in step S1 to obtain multi-dimensional traffic flow data;

S3. Short-term traffic flow model training, using long-term short-term memory network to train traffic flow data;

S4. Prediction of short-term traffic flow model. Based on the trained short-term traffic flow model, long-short-term memory network is used to predict short-term traffic flow of multiple road sections under complex road network.
The short-term traffic flow forecasting method according to claim 1, wherein step S1 comprises:

S11. Select N road sections, where N is a positive integer ≥ 2, and the road sections include intersections and/or road section points;

S12. Collect the traffic flow data of the N road sections for the selected road sections through ground induction coil triggering, video triggering or GPS data.
The short-term traffic flow forecasting method according to claim 2, wherein step S2 comprises:

S21. Perform data cleaning on the traffic flow data collected in step 1;

S22. Perform time slicing on the cleaned data at a set time interval t, and cut the data of each road segment into M time segments, so as to generate an N*M traffic flow data matrix.
The short-term traffic flow forecasting method according to claim 3, wherein step S3 comprises:

S31. Traffic flow data preprocessing, normalize the traffic flow data obtained in step S2, the normalized traffic flow x i ' at the current moment is:

In the formula, x′ i is the normalized traffic flow at the current moment, x max is the maximum value of the traffic flow at the historical moment, and x min is the minimum value of the traffic flow at the historical moment;

S32. Input the normalized N*M traffic flow data matrix as a training sample into the long-short-term memory network model for training.
The short-term traffic flow forecasting method according to claim 4, wherein step S32 comprises:

S321, the matrix extracts advanced features through the convolution layer;

S322, pooling layer processing, to reduce the number of parameters and control overfitting;

S323, activation function processing, using ReLU activation function processing to avoid mode collapse;

S324, Dropout layer processing, to alleviate the occurrence of overfitting;

S325. Based on the attention mechanism, learn the associated information of the time series data through the LSTM layer of the long short-term memory network to improve the model training effect;

S326. Evaluate the prediction effect of the model training on the data through a loss function.
The short-term traffic flow prediction method according to claim 5, wherein the long-short-term memory network LSTM layer processing based on the attention mechanism of step S325 comprises:

The output o of the long-term short-term memory network LSTM is weighted and summed with the layer weight W, and the attention coefficient a is obtained through softmax, and a is mapped to the output value through the fully connected layer:

a＝softmax(w T sigmoid(o))……………………………Formula 2

output=sigmoid(w T (oa T ))………………………………………Formula 3 In the formula, o is the output of LSTM, and W is the weight of the first layer.
According to the short-term traffic flow prediction method described in claim 5 or 6, it is characterized in that the loss function of step S326 adopts the mean square error loss, and the statistical parameter is the mean value of the sum of squares of the corresponding point errors of the predicted data and the original data, calculated The formula is:

In the formula, y i represents the true value of the i-th sample;

f( xi ) represents the predicted value of the i-th sample;

n is the number of samples.
The short-term traffic flow prediction method according to claim 7, wherein the short-term traffic flow model prediction of step S4 comprises:

S41. Collect the traffic flow data of the previous M time periods, and perform data normalization processing;

S42. Input the normalized traffic flow data matrix into the long-short-term memory network model to predict the short-term traffic flow of multiple road sections.
A computer-readable storage medium, on which computer-executable program instructions are stored, is characterized in that: the instructions are executed by a processor and a memory, so as to realize the method described in any one of claims 1-8.
A short-term traffic flow prediction system under a complex road network is characterized in that: the system includes a data acquisition module connected by telecommunication, a data preprocessing module, a short-term traffic flow training prediction module, a prediction effect evaluation module and a traffic flow display module, in,

The data acquisition module includes a data acquisition unit and a timer; the data acquisition unit includes a ground sensing coil, a video monitor or a GPS assembly, and under the control of the timer, the acquisition unit synchronously collects a plurality of road sections at equal time intervals traffic flow data;

The data preprocessing module receives the traffic flow data of the data acquisition module, and performs data cleaning and time slicing to obtain a traffic flow data matrix;

The short-term traffic flow training prediction module includes a long-short-term memory network training unit and a prediction unit based on the attention mechanism, the training unit is used to train the historical traffic flow data matrix, and the prediction unit receives the current traffic flow data matrix Predict the traffic flow data in the next period;

The prediction effect evaluation module evaluates the accuracy of the short-term traffic flow training prediction module through a loss function;

The traffic flow display module visually displays historical data, current real data, forecast data and forecast effects of multiple road sections.