WO2023273724A1

WO2023273724A1 - Traffic analysis method, related device, and readable storage medium

Info

Publication number: WO2023273724A1
Application number: PCT/CN2022/095206
Authority: WO
Inventors: 蒋鑫; 纪雅琪; 王健; 童恒金; 杜豫川; 都州扬; 潘宁; 刘成龙; 曾俊益; 曾程; 敖星冉; 吴荻非
Original assignee: 中移(上海)信息通信科技有限公司; 中移智行网络科技有限公司; 中国移动通信集团有限公司
Priority date: 2021-06-29
Filing date: 2022-05-26
Publication date: 2023-01-05
Also published as: CN113256986B; CN113256986A

Abstract

A traffic analysis method, a related device, and a readable storage medium, relating to the field of traffic. The method comprises: step 101, generating a first tensor according to traffic data detected by detectors in a target road network within a first duration, wherein the length of a spatial dimension of the first tensor is S road segments in the target road network, the length of a temporal dimension is T time points within the first duration, the length of a traffic data dimension is R types of traffic data detected by the detectors, S and T are both positive integers, and R is an integer greater than 1; step 102, fusing the spatial dimension and the traffic data dimension in the first tensor by using an attention mechanism to obtain a second tensor; and step 103, determining target information of the target road network according to a third tensor, wherein the third tensor is the second tensor or is determined on the basis of the second tensor, and the target information comprises at least one of the following: a traffic mode, a traffic pattern, and a future traffic state. By performing traffic analysis using multi-dimensional traffic data, the accuracy of traffic analysis is improved.

Description

Traffic analysis method, related equipment and readable storage medium

Cross References to Related Applications

This application claims the priority of the Chinese patent application with the application number 202110727130.5 and the application title "traffic analysis method, related equipment and readable storage medium" submitted to the China Patent Office on June 29, 2021, the entire contents of which are incorporated by reference in this application.

technical field

The embodiments of the present application relate to the traffic field, and in particular, to a traffic analysis method, related equipment, and a readable storage medium.

Background technique

With the rapid development of the transportation field, traffic congestion is increasing day by day. In order to ensure that the traffic management measures are reasonable and effective, it is necessary to obtain the traffic data on the road network, and use the obtained traffic data to analyze the traffic. In the prior art, only a single traffic data is used for traffic analysis, resulting in low accuracy of traffic analysis.

Contents of the invention

Embodiments of the present application provide a traffic analysis method, related equipment, and a readable storage medium, so as to solve the problem of low traffic analysis accuracy in the prior art.

In order to solve the above problems, the application is implemented as follows:

In the first aspect, the embodiment of the present application provides a traffic analysis method, the method comprising:

According to the traffic data detected by each detector in the target road network within the first time length, a first tensor is generated, the length of the space dimension of the first tensor is S road sections in the target road network, and the time dimension The length is T time points in the first duration, the length of the traffic data dimension is the R type traffic data detected by the detector, S and T are both positive integers, and R is an integer greater than 1;

Using an attention mechanism to fuse the spatial dimension and traffic data dimension in the first tensor to obtain a second tensor;

Determine the target information of the target road network according to the third tensor;

Wherein, the third tensor is the second tensor or is determined based on the second tensor; the target information includes at least one of the following: traffic mode, traffic law, and future traffic state.

In the second aspect, the embodiment of the present application also provides a traffic analysis device, including:

A generating module, configured to generate a first tensor according to the traffic data detected by each detector in the target road network within the first duration, the length of the spatial dimension of the first tensor being S in the target road network sections, the length of the time dimension is T time points in the first duration, the length of the traffic data dimension is the R type traffic data detected by the detector, S and T are both positive integers, and R is greater than 1 integer;

The fusion module is used to fuse the spatial dimension and the traffic data dimension in the first tensor by using the attention mechanism to obtain the second tensor;

The first determination module is configured to determine the target information of the target road network according to the third tensor;

In the third aspect, the embodiment of the present application also provides an electronic device, including: a transceiver, a memory, a processor, and a program stored in the memory and operable on the processor; the processor is used to Reading the program in the memory implements the method described in the aforementioned first aspect.

In a fourth aspect, the embodiment of the present application further provides a readable storage medium for storing a program, and when the program is executed by a processor, the method described in the aforementioned first aspect is implemented.

In the embodiment of the present application, the electronic device performs spatio-temporal splicing on the multi-dimensional traffic data detected by the detector to obtain the first tensor; after that, it performs fusion processing on the spatial dimension and the traffic data dimension in the first tensor to obtain the first tensor Two tensors, and then according to the second tensor, determine the target information of the target road network, the target information includes at least one of the following: traffic mode, traffic law and future traffic state. It can be seen that the electronic device in the embodiment of the present application makes full use of the multi-dimensional traffic data detected by the detector to analyze the traffic of the target road network, thereby improving the accuracy of the traffic analysis.

Description of drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that need to be used in the description of the embodiments of the present application will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application. For those skilled in the art, other drawings can also be obtained based on these drawings without any creative effort.

Fig. 1 is a schematic flow chart of the traffic analysis method provided by the embodiment of the present application;

Fig. 2 is a schematic diagram of the first tensor provided by the embodiment of the present application;

Fig. 3 is a schematic diagram of feature extraction of the tensor provided by the embodiment of the present application;

FIG. 4 is a schematic diagram of the fusion of tensors provided by the embodiment of the present application;

Fig. 5 is a schematic diagram of the first model provided by the embodiment of the present application;

Fig. 6 is a schematic structural diagram of the traffic analysis device provided by the implementation of the present application;

FIG. 7 is a schematic structural diagram of an electronic device provided by the implementation of the present application.

detailed description

The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of them. Based on the embodiments in this application, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the protection scope of this application.

The terms "first", "second" and the like in the embodiments of the present application are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. Furthermore, the terms "comprising" and "having", as well as any variations thereof, are intended to cover a non-exclusive inclusion, for example, a process, method, system, product or device comprising a sequence of steps or elements is not necessarily limited to the expressly listed instead, may include other steps or elements not explicitly listed or inherent to the process, method, product or apparatus. In addition, the use of "and/or" in this application means at least one of the connected objects, such as A and/or B and/or C, means that A alone, B alone, C alone, and both A and B exist, Both B and C exist, both A and C exist, and there are 7 situations where A, B, and C all exist.

The traffic analysis method provided by the embodiment of the present application will be described below.

The traffic analysis method in the embodiment of the present application may be executed by an electronic device. In practical applications, electronic equipment can be servers, mobile phones, tablet computers (Tablet Personal Computer), laptop computers (Laptop Computer), personal digital assistants (Personal Digital Assistant, PDA), mobile Internet devices (Mobile Internet Device, MID ), wearable device (Wearable Device) or vehicle-mounted device, etc.

Referring to FIG. 1 , FIG. 1 is a schematic flowchart of a traffic analysis method provided in an embodiment of the present application. As shown in Figure 1, the following steps may be included:

Step 101, according to the traffic data detected by each detector in the target road network within the first time length, generate a first tensor, the length of the spatial dimension of the first tensor is S road sections in the target road network , the length of the time dimension is T time points within the first duration, and the length of the traffic data dimension is the R-type traffic data detected by the detector. Both S and T are positive integers, and R is an integer greater than 1.

In practical applications, the detectors are arranged on each lane of each road section of the target road network. Optionally, one detector may be arranged on one lane of one road section, that is, one detector may be used to detect traffic data of one lane of one road section.

Each detector can detect the value of R-type traffic data according to a preset frequency, and the R-type traffic data can include but not limited to the average speed of each lane, occupancy rate, number of driving vehicles, lane occupancy time, etc., which can reflect the traffic state data . In the embodiment of the present application, the R-type traffic data may also be referred to as the R-type traffic state.

In this step, the electronic device can obtain the overall traffic status of the target network within the first period of time by acquiring the traffic data detected by each detector in the target network within the first period of time, and then can analyze the target network in the first period of time. Perceive the traffic state within the first duration, and use the perceived traffic state to predict the future traffic state of the target network. The first preset duration can be set in advance according to demand, such as 1 hour or 1 day Wait.

After the electronic device acquires the traffic data detected by each detector in the target network within the first time period, it can generate a first tensor, the first tensor is a multi-dimensional space-time tensor of traffic data, and the first The tensor includes all traffic data detected by the detectors of each road section in the target road network within the first time period.

It can be understood that, when the target network includes S road sections, the detector performs T detections within the first time period, and the detector detects R traffic data, the first tensor is S×T ×R tensor, where S represents the length of the spatial dimension; T represents the length of the time dimension; R represents the length of the traffic data dimension.

Considering that the millimeter-wave radar has the following advantages: it is less affected by the environment, such as haze, rain, snow and light; the detection coverage is large, the viewing angle can reach 120°, and the detection distance can reach 250 meters; the engineering is simple; Additional computing power requirements; low maintenance costs. The detector in this embodiment of the present application may be a millimeter-wave radar detector. In this way, the traffic data detected by the detector is more accurate, thereby further improving the accuracy of traffic analysis. Of course, it can be understood that the detector in the embodiment of the present application may also be other types of detectors, such as a coil detector or a geomagnetic detector, which may be determined according to actual needs, which is not limited in the embodiment of the present application.

Step 102, using the attention mechanism to fuse the spatial dimension and traffic data dimension in the first tensor to obtain the second tensor.

In the embodiment of the present application, after the electronic device generates the first tensor, it can also use the attention mechanism to capture and identify the regional features of the spatial dimension and traffic data dimension in the first tensor, so as to improve the recognition and capture of regional features the accuracy. Since the more accurate the feature recognition and capture, the better the analysis effect of the traffic state. Therefore, the analysis of the traffic state by using the second tensor obtained by fusion can further improve the accuracy of the traffic state analysis.

Step 103, according to the third tensor, determine the target information of the target road network.

During specific implementation, in the first implementation manner, the electronic device may directly determine the target information of the target road network based on the second tensor obtained through fusion. In the second implementation manner, the electronic device may further detect whether there is a vacant value in the second tensor, and if there is a vacant value in the second tensor, fill in the second tensor, and use the first Three tensors, which determine the target information of the target road network. It can be seen that, compared with the first implementation manner, the accuracy of the target information of the target road network determined by the second implementation manner is higher.

In this embodiment of the present application, the traffic mode may be characterized by a degree of congestion, and the degree of congestion in different traffic modes is different. The traffic law can be represented by the average value of various traffic data in the target time or target space. The future traffic state can be represented by multi-dimensional traffic data space-time tensor.

In the traffic analysis method of the embodiment of the present application, the electronic device performs spatiotemporal splicing on the multi-dimensional traffic data detected by the detector to obtain the first tensor; after that, fusion processing is performed on the spatial dimension and the traffic data dimension in the first tensor, The second tensor is obtained, and then the target information of the target road network is determined according to the second tensor, and the target information includes at least one of the following: traffic mode, traffic law, and future traffic state. It can be seen that the electronic device in the embodiment of the present application makes full use of the multi-dimensional traffic data detected by the detector to analyze the traffic of the target road network, thereby improving the accuracy of the traffic analysis.

In the embodiment of the present application, optionally, generating the first tensor according to the traffic data detected by each detector in the target road network within the first duration includes:

Generate a time sequence corresponding to T time points detected by each detector in the target road network, and a space sequence corresponding to S road sections in the target road network;

Concatenating the value of the R-type traffic data corresponding to each time point in the time series and the value of the R-type traffic data corresponding to each road segment in the space series to obtain the first tensor.

In this optional implementation manner, after the electronic device acquires the traffic data detected by each detector in the target network within the first duration, it may perform spatiotemporal stitching on the acquired traffic data to generate the first tensor.

In actual implementation, the value of traffic data is highly correlated with the time dimension. After each unit of time, the value of each traffic data will change due to reasons such as vehicles entering, leaving or parking and waiting, accident collisions, etc. in each lane of the road network. Therefore, the electronic device may generate a time sequence according to the time sequence detected by the detector, and the time sequence includes the T time points.

Traffic data such as speed and occupancy of the lanes will change due to the alternate entry of vehicles, or the interaction between front and rear vehicles, and left and right vehicles between various road sections of the same road network and between lanes of the same road section. Therefore, the electronic device may generate a spatial sequence according to the spatial sequence of the S road sections, where the spatial sequence includes the S road sections.

After generating the time series and the space series, the electronic device forms a grid map with time information and spatial relative relationships based on time and space dimensions. R traffic data can be represented by R grid graphs, and the R grid graphs form a multi-dimensional traffic state space-time tensor according to the time-space correspondence, that is, the first tensor, where each grid graph represents a class For traffic data, each grid in the grid graph represents the value of a type of traffic data for a lane at a point in time. For easy understanding, please refer to FIG. 2 . In FIG. 2 , the traffic data dimension of the first tensor is 3, but this does not limit the value of the traffic data dimension.

Through the above method, the temporal and spatial splicing of the traffic data of the target road network within the first time length can be realized, and the first tensor can be obtained. In this way, the traffic state obtained by using the first tensor analysis can be more in line with the actual traffic state, which can improve the accuracy of traffic analysis.

In the embodiment of the present application, optionally, the use of the attention mechanism to fuse the spatial dimension and the traffic data dimension in the first tensor to obtain the second tensor includes:

Extracting the regional features of the spatial dimension and the traffic data dimension in the first tensor to obtain an intermediate tensor;

The intermediate tensor is input into the attention fusion network to obtain the weight value of each regional feature in the intermediate tensor;

Multiplying the weight value of each region feature with the first tensor to obtain a second tensor.

During specific implementation, the electronic device may use a feature extraction model to extract regional features of the spatial dimension and the traffic data dimension in the first tensor. The feature extraction model can be a convolutional neural network, but is not limited thereto. The convolutional neural network can use convolution kernels with different weights to extract the regional features of the input tensor. Convolution kernels of different sizes can capture regional features of different scales.

As shown in Figure 3, in the feature extraction model, the first tensor is filtered and calculated by multiple convolution kernels to obtain the intermediate tensor π, and the size of the intermediate tensor π is H×W×T, where H represents the spatial dimension W represents the length of the traffic data dimension, and T represents the length of the time dimension.

The electronic device may input the intermediate tensor π into the attention fusion network to obtain the weight value ε of each region feature in the intermediate tensor. As shown in Figure 4, the attention fusion network can average each plane of the intermediate tensor π, compress 1×1×T, and obtain the weight value ε of each region feature in the intermediate tensor.

Afterwards, as shown in FIG. 4 , the electronic device may multiply the weight value ε of each region feature by the first tensor to obtain a second tensor of depth fusion.

Through the above method, the electronic device can assign different weights to different regional features of the first tensor, and then fuse them to obtain a multi-dimensional traffic data tensor with an attention mechanism, thereby further improving the accuracy of traffic analysis.

In the embodiment of the present application, optionally, the attention mechanism is used to fuse the spatial dimension and traffic data dimension in the first tensor, and after obtaining the second tensor, the third tensor is used to determine the Before the target information of the target road network, the method also includes:

detecting whether there is a missing value in the second tensor;

If there is a vacant value in the second tensor, the vacant value is filled by using a weight-based interpolation method to obtain a third tensor.

In this optional implementation, it is considered that the detector may lack data collection due to at least one of the following factors: the detector is interfered by obstructions, buildings or bad weather; the detector is debugged, the built-in program has errors, and the like. In order to improve the accuracy of traffic analysis, missing data can be filled. Filling in the gaps can be to use the collected traffic data to perform numerical interpolation or reasonable prediction in the missing places to make the data reasonable and complete.

In specific implementation, vacant values can be filled in the following ways:

1) Search for the vacant value of each dimension layer, and locate the position of the vacant value. Given the distribution of a traffic data variable ri∈R on the spatio-temporal layer:

C＝{(s ₁ ,t ₁ ,r _i ),(s ₁ ,t ₂ ,r _i ),…(s ₁ ,t _k-1 ,r _i ),(s ₁ ,t _k+1 ,r _i )…(s _m ,t _n ,r _i )}

The judging vacancy value is c=(s ₁ , t _k , _ri ), which means that the value of traffic data ri is missing at time _s ₁ in space t _k .

2) Use the weight-based interpolation method to fill in the vacant values. The vacant value is related to the state variable values of the front and back time series and the left and right space series. Therefore, the weighted method is used to find the expectation of the variables at the front, back, left, and right sides of the variable, so as to fill the vacant value at the center.

c＝α ₁ (s ₀ ,t _k-1 , _ri )+α ₂ (s ₀ ,t _k+1 , _ri )+α ₃ (s ₂ ,t _k-1 , _ri )+α ₄ ( s ₂ ,t _k+1 ,r _i )

Σα ₁ +α ₂ +α ₃ +α ₄ =1

Among them, the weighting coefficients α ₁ , α ₂ , α ₃ , and α ₄ depend on the extent to which the vacancy value is affected by surrounding spatiotemporal variables, and can be specifically set according to actual needs, which is not limited in this embodiment of the present application. Generally, the greater the impact of surrounding variables on the traffic state value at the vacancy, the higher the weighting coefficient.

It should be noted that the above calculation formula of c is only an example, and does not limit the calculation method of c. For example, c can also be calculated by the following formula:

c＝α ₁ (s ₁ ,t _k-1 ,r _i )+α ₂ (s ₁ ,t _k+1 ,r _i )+α ₃ (s ₀ ,t _k ,r _i )+α ₄ (s ₂ ,t _k ,r _i )

The determination of the target information of the target road network in the embodiment of the present application will be described below.

1. The target information includes traffic patterns.

In this case, optionally, the determining the target information of the target road network according to the third tensor includes:

Perform tensor decomposition on the third tensor to obtain the first factor matrix corresponding to the time dimension, the second factor matrix corresponding to the space dimension, and the third factor matrix corresponding to the traffic data;

The traffic mode of the target road network is determined according to the first factor matrix, the second factor matrix and the third factor matrix.

In this optional implementation manner, the electronic device uses tensor (Canonical Polyadic, CP) decomposition to determine the traffic mode. It should be noted that, in other implementation manners, the electronic device may also use the third tensor to determine the traffic mode in other ways, for example, the electronic device may determine the traffic mode by using the second model, and the input of the second model is the The third tensor is output as the first factor matrix, the second factor matrix, and the third factor matrix. The embodiment of the present application does not limit the way of determining the traffic mode according to the third tensor.

Traffic mode, also known as traffic travel mode or traffic behavior mode, reflects the inherent similarity between the traffic states of individuals or road segments within a given time and space range. In practical applications, individuals or road segments with significant similar connections can be classified into the same traffic mode.

Assume that the length of the spatial dimension of the third tensor is n road sections, the length of the time dimension is m time points, the length of the traffic data dimension is i-type traffic data, n and m are both positive integers, and i is greater than 1 an integer of .

Then, the size of the first factor matrix A corresponding to the time dimension obtained by performing CP decomposition on the third tensor is m×Q, and the size of the second factor matrix B corresponding to the space dimension is n×Q. The size of the third factor matrix C corresponding to the data is i×Q, that is, the first factor matrix is an m×Q matrix, the second factor matrix B is an n×Q matrix, and the third factor matrix C is i ×Q matrix. Wherein, Q represents the number of traffic modes of the target road network, and Q is a positive integer.

Afterwards, the electronic device may determine the traffic mode of the target road network based on the first factor matrix, the second factor matrix and the third factor matrix obtained through decomposition.

Optionally, the determining the traffic mode of the target road network according to the first factor matrix, the second factor matrix and the third factor matrix includes:

Obtaining Q groups of traffic data sequences according to the first factor matrix, the second factor matrix and the third factor matrix;

Acquiring Q traffic modes corresponding to the Q group of traffic data sequences.

During specific implementation, the electronic device may sequentially extract and expand the obtained first factor matrix A, second factor matrix B, and third factor matrix C according to row or column q=1, 2, 3, ..., Q, A Q group of traffic data sequences (also called a travel mode sequence or a state sequence) can be obtained. Each group of travel modes includes three vectors of m×1, n×1, and i×1, and each group of travel modes can be used to determine a traffic mode, so that Q groups of traffic data sequences can reveal Q types of traffic with different characteristics traffic pattern.

For the solution of the first factor matrix A, the second factor matrix B and the third factor matrix C, it can be realized in the following way:

constructing an objective function that reflects the difference between a first value that is a value of the third tensor and a second value that is based on the first factor matrix , the second factor matrix and the third factor matrix are calculated;

optimizing the first factor matrix, the second factor matrix and the third factor matrix according to the objective function;

The traffic mode of the target road network is determined according to the optimized first factor matrix, second factor matrix and third factor matrix.

Note that the third tensor is X, and the relationship between X and the first factor matrix A, the second factor matrix B and the third factor matrix C can be expressed by the following formula:

The relationship between the elements in X and the elements in the first factor matrix A, the second factor matrix B and the third factor matrix C can be expressed by the following formula:

Construct the objective function f, the objective function f can reflect X and the second value

difference between. Optionally, the objective function f can be expressed by the following formula:

The objective function f can be simplified as:

To optimize the factor matrix, the gradient descent method can be used to solve the optimization decision variables a _mr , b _nr , c _ir . Optionally, partial derivatives can be obtained for each decision variable first, and each decision variable can be updated using the partial derivative value, and iteratively updated until the objective function converges, that is, the optimum is achieved, that is, the difference between the first value and the second value is the smallest , the second value is infinitely close to the first value. It can be understood that when the objective function is optimized, each factor matrix can be optimized, and the optimal first factor matrix, second factor matrix and third factor matrix can be used to determine the traffic mode of the target road network , so that the accuracy of traffic mode determination can be further improved.

The update of each decision variable can be achieved by the following formula:

2. The target information includes traffic rules.

The spatio-temporal law of traffic is reflected in a given space-time range, among individual travel individuals or travel sections, individuals or sections present a similar or identical state in a specific time period and space. The traffic law can be represented by the average value of various traffic data in the target time or target space.

In this case, optionally, the determining the target information of the target road network according to the third tensor includes at least one of the following:

1) Obtain the average value of the kth traffic data of the target road network within the target duration;

2) Obtain the average value of the kth traffic data in the target space of the target road network within the first duration;

Wherein, the time unit of the target duration includes at least one of the following: day; week; year; the spatial unit of the target space includes at least one of the following: lane; road section; road network; the value range of k is 1 to i .

For 1), the electronic device can perform daily and/or weekly and/or annual traffic status analysis.

On the time scale, the hours detected by the detector, seven days a week, and December a year are used as time units (also called time scales or aggregate units) to collect statistics on various types of traffic data:

A＝{a ₁ ,a ₂ ,a ₃ ,a ₄ ,…a _k }

Among them, a _k represents the average value of the i-th traffic state of the target road network within the (o,p) time period.

When the time scale is day, and the continuous statistical time of (o,p) is 1 hour, the state values of all 24 hours in a day are calculated to obtain the daily traffic state analysis table.

When the time scale is a week, and the continuous statistical time of (o,p) is 1 day, the state values of all 7 days in a week are calculated to obtain a weekly traffic state analysis table.

When the time scale is year, the continuous statistical time of (o,p) is 1 month, and the state values of all 12 months in a year are calculated to obtain the annual traffic state analysis table.

For 2), the electronic device can perform lane-level and/or section-level and/or road-network-level traffic state analysis.

On the spatial scale, the lanes, road sections, and road networks detected by the detector are used as aggregate units to make statistics on various traffic states:

B＝{b ₁ ,b ₂ ,b ₃ ,b ₄ ,…b _k }

Among them, b _k represents the average value of the i-th traffic state on the (q, u) section at time t _n . The above formula can be used to obtain the traffic status table of a certain lane level, a certain road section or a certain road network level in a city at a certain moment.

3. The target information includes future traffic status.

Based on the third tensor, it is possible to predict the traffic state in the future. Traffic state prediction, reflecting the prediction of various traffic states in the road sections detected by the millimeter-wave radar within a given time range in the future. The multi-dimensional traffic data tensor established based on the data detected by the detector has strong temporal and spatial dependencies, and can make accurate predictions of future traffic conditions by using its tight temporal and spatial dependencies.

Optionally, the determining the target information of the target road network according to the third tensor includes:

Input the third tensor into the trained first model, and predict the fourth tensor, which is used to reflect the traffic state of the target road network within the second time period, and the second The duration is the next duration of the first duration;

Wherein, the first hidden state when the first model predicts the fourth tensor is determined by the third tensor and the historical hidden state, and the historical hidden state is the hidden state of the first model in the third duration State determination, the third duration is a previous duration of the first duration; the hidden state of the first model is used to determine the output of the first model.

In this optional implementation manner, the first model is used to predict the traffic state at a future moment, the input of the first model is the third tensor, and the output is the tensor of the future duration corresponding to the third tensor , which can be used to reflect the traffic status of the target road network within a narrow time period.

The structure of the first model can be seen in Fig. 5, and the first model predicts the future traffic state and can comprise following three steps:

1) Enter the variables.

For the convenience of understanding, the above-mentioned duration is regarded as a moment in the following, and the future duration is the future moment. Sorting and sorting the variables according to the time series, input x ⁽¹⁾ , x ⁽²⁾ ,…, x ^(t) in order, respectively representing the third tensor calculated by the electronic device at time t. The other variables of the predictive model are:

h ^(t) represents the hidden state of the model at time t, and h ^(t) is jointly determined by the input x ^(t) at the corresponding time and the hidden state h ^(t-1) at the previous moment.

o ^(t) represents the output of the model at time t, and o ^(t) is only determined by the model's current hidden state h ^(t) .

L ^(t) represents the loss function of the model at time t.

y ^(t) represents the true value at time t.

U, V, W represent the shared weight matrix in the prediction model.

2) Forward propagation calculation.

Based on the above specified variables, the forward propagation calculation of the model is performed. For any moment t, the hidden state h ^(t) at this moment is obtained from the multi-dimensional traffic state tensor x ^(t) at this moment and the hidden state h ^(t-1) at the previous moment:

h ^(t) = tan(z ^(t) ) = tan(Ux ^(t) +Wh ^(t-1) +b)

By performing two shared weight matrix operations on the hidden variables at time t, the estimated value of the multi-dimensional traffic state at this time is obtained

o ^(t) = Vh ^(t) +c

3) Backpropagation optimization.

In order to achieve an accurate prediction of the future traffic state, it is necessary to weigh the estimated value of the traffic state obtained at the previous moment

and the real traffic state value y ^(t) to quantify the difference, and establish a loss function (also called a difference function) L between the two.

Optionally, before the third tensor is input into the trained first model, and before the fourth tensor is predicted, the method further includes:

Get the sample tensor;

inputting the sample tensor into the untrained first model to obtain a fifth tensor;

determining a loss function corresponding to the first model according to the sample tensor and the fifth tensor;

According to the loss function, the weight value of the first model is adjusted to obtain the trained first model.

The loss function can be calculated by the following formula:

The gradient descent algorithm is used to calculate the partial derivatives of the decision variables v and V of the difference function L, iteratively calculate multiple rounds, and continuously update c and V to optimize the model and achieve accurate predictions.

The partial derivatives of the decision variables U, W, and b of the difference function L are gradually calculated by the stepwise gradient descent method, and iterative for multiple rounds, and the model is continuously updated to optimize the model and achieve accurate prediction.

Gradually update W:

UPDATE U:

Update b:

When training the model, the model parameters are iteratively optimized based on the loss function, and when the loss function is smaller than the specified threshold, no update is performed. Input the multi-dimensional traffic state variable at this moment into the model, and after calculation, the multi-dimensional traffic state variable corresponding to the space at the next moment can be obtained, that is, the prediction result can be obtained.

The various optional implementation manners introduced in the embodiments of the present application may be implemented in combination with each other if they do not conflict with each other, or may be implemented independently, which is not limited in the embodiments of the present application.

The embodiment of this application includes the following contents:

The electronic equipment collects the detection data of each millimeter-wave radar detection point on the target road network. According to the matching results between the stake number of the millimeter-wave radar and the map positioning, the continuous millimeter-wave radar data sets in the road network are spliced in time and space dimensions to restore the traffic status at the lane level. Among them, each data tensor contains no less than two types of traffic status information, including but not limited to information such as average speed, occupancy rate, and number of vehicles in each lane in the road network.

The electronics output spliced and fused mmWave radar data. According to the millimeter-wave radar multi-source data fusion algorithm of the present application, the lane-level data tensor formed by splicing is deeply fused to obtain the millimeter-wave radar multi-dimensional traffic state tensor fused with multi-source information.

The millimeter-wave radar data formed by the fusion of electronic equipment.

The electronic device outputs the predicted traffic state value of the lane in the future.

The embodiment of the present application has at least the following beneficial effects:

This application adopts an effective data format for millimeter-wave radar data, and proposes an effective urban road network lane-level algorithm based on this data format, which can not only realize the travel mode analysis of traffic road network data, but also can analyze Interval specified roads for accurate traffic state prediction.

This application adopts the method of accurately splicing all the time range and area range data covered by the millimeter wave radar to realize the deep fusion of multi-source data generated by the millimeter wave radar, improve the accuracy of perception and prediction; and process the missing data , to adapt to the data problems existing in the real scene, and lay the foundation for the further use of millimeter-wave radar data in the future.

The data analysis and prediction method of the multi-dimensional millimeter-wave radar proposed in this application can adapt to the transmission of traffic roadside equipment in the 5G scenario, and can also be oriented to distributed computing and processing of data under edge computing. This proposal can not only realize the travel mode analysis of traffic road network data in the existing technical environment, but also adapt to the rapid processing and calculation of traffic data in the future emerging technology scenarios.

Referring to FIG. 6, FIG. 6 is one of the structural diagrams of the traffic analysis device provided by the embodiment of the present application. As shown in Figure 6, the traffic analysis device 600 includes:

The generating module 601 is configured to generate a first tensor according to the traffic data detected by each detector in the target road network within the first duration, and the length of the spatial dimension of the first tensor is S road sections, the length of the time dimension is T time points in the first duration, the length of the traffic data dimension is the R type traffic data detected by the detector, S and T are both positive integers, and R is greater than 1 an integer of

Fusion module 602, for utilizing the attention mechanism to fuse the spatial dimension and the traffic data dimension in the first tensor to obtain the second tensor;

The first determination module 603 is configured to determine the target information of the target road network according to the third tensor;

Optionally, the generating module 601 includes:

A generation submodule is used to generate a time sequence corresponding to T time points detected by each detector in the target road network, and a space sequence corresponding to S road sections in the target road network;

The splicing sub-module is used to splice the value of the R-type traffic data corresponding to each time point in the time series and the value of the R-type traffic data corresponding to each road segment in the space sequence to obtain the first tensor.

Optionally, the fusion module 602 includes:

The extraction submodule is used to extract the regional features of the spatial dimension and the traffic data dimension in the first tensor to obtain an intermediate tensor;

The first acquisition submodule is used to input the intermediate tensor into the attention fusion network to obtain the weight value of each region feature in the intermediate tensor;

The second acquisition sub-module is configured to multiply the weight value of each region feature by the first tensor to obtain a second tensor.

Optionally, the traffic analysis device 600 also includes:

A detection module, configured to detect whether there is a vacant value in the second tensor;

The first obtaining module is configured to, in the case that there is a vacant value in the second tensor, use a weight-based interpolation method to fill in the vacant value to obtain a third tensor.

Optionally, when the target information includes a traffic mode, the first determining module 603 includes:

The decomposition sub-module is used to perform tensor decomposition on the third tensor to obtain the first factor matrix corresponding to the time dimension, the second factor matrix corresponding to the space dimension, and the third factor matrix corresponding to the traffic data;

A determining submodule, configured to determine the traffic mode of the target road network according to the first factor matrix, the second factor matrix and the third factor matrix.

Optionally, the determining submodule includes:

a construction unit, configured to construct an objective function, the objective function is used to reflect the difference between a first value and a second value, the first value is the value of the third tensor, and the second value is based on the The first factor matrix, the second factor matrix and the third factor matrix are calculated;

an optimization unit, configured to optimize the first factor matrix, the second factor matrix, and the third factor matrix according to the objective function;

The determining unit is configured to determine the traffic mode of the target road network according to the optimized first factor matrix, second factor matrix and third factor matrix.

Optionally, the length of the spatial dimension of the third tensor is n road sections, the length of the time dimension is m time points, the length of the traffic data dimension is i-type traffic data, n and m are both positive integers, i is an integer greater than 1;

The first factor matrix is an m×Q matrix, the second factor matrix is an n×Q matrix, the third factor matrix is an i×Q matrix, and Q is a positive integer;

The determination submodule includes:

A first acquisition unit, configured to obtain Q groups of traffic data sequences according to the first factor matrix, the second factor matrix and the third factor matrix;

The second acquiring unit is configured to acquire Q traffic modes corresponding to the Q group of traffic data sequences.

Optionally, the length of the traffic data dimension of the third tensor is traffic data of type i, where i is an integer greater than 1;

When the target information includes traffic rules, the first determination module 603 is used for at least one of the following:

Obtain the average value of the kth traffic data of the target road network within the target duration;

Obtaining the average value of the kth type of traffic data in the target space of the target road network within the first time period;

Optionally, when the target information includes the future traffic state, the first determination module 603 is configured to:

Optionally, the traffic analysis device 600 also includes:

The second acquisition module is used to acquire the sample tensor;

A third acquisition module, configured to input the sample tensor into the untrained first model to obtain a fifth tensor;

A second determining module, configured to determine a loss function corresponding to the first model according to the sample tensor and the fifth tensor;

The adjusting module is used to adjust the weight value of the first model according to the loss function to obtain the trained first model.

The traffic analysis device 600 can realize various processes of the method embodiment in FIG. 1 in the embodiment of the present application, and achieve the same beneficial effect. To avoid repetition, details are not repeated here.

The embodiment of the present application also provides an electronic device. Referring to FIG. 7, the electronic device may include a processor 701, a memory 702, and a program 7021 stored in the memory 702 and operable on the processor 701. When the program 7021 is executed by the processor 701, the method embodiment corresponding to FIG. 1 may be implemented. Any steps in the method and achieving the same beneficial effect will not be repeated here.

Those skilled in the art can understand that all or part of the steps for implementing the methods of the above embodiments can be completed by program instructions related hardware, and the program can be stored in a readable medium. The embodiment of the present application also provides a readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, any step in the above-mentioned method embodiment corresponding to FIG. 1 can be implemented, and The same technical effect can be achieved, so in order to avoid repetition, details will not be repeated here.

The storage medium is, for example, a read-only memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), a magnetic disk or an optical disk, and the like.

The above is the preferred implementation of the embodiment of the present application. It should be pointed out that for those of ordinary skill in the art, without departing from the principles described in the present application, some improvements and modifications can also be made. These improvements and Retouching should also be regarded as the protection scope of this application.

Industrial Applicability

Claims

A traffic analysis method comprising:

According to the traffic data detected by each detector in the target road network within the first time length, a first tensor is generated, the length of the space dimension of the first tensor is S road sections in the target road network, and the time dimension The length is T time points in the first duration, the length of the traffic data dimension is the R type traffic data detected by the detector, S and T are both positive integers, and R is an integer greater than 1;

Using an attention mechanism to fuse the spatial dimension and traffic data dimension in the first tensor to obtain a second tensor;

Determine the target information of the target road network according to the third tensor;

Wherein, the third tensor is the second tensor or is determined based on the second tensor; the target information includes at least one of the following: traffic mode, traffic law, and future traffic state.
The method according to claim 1, wherein said generating the first tensor according to the traffic data detected by each detector in the target road network within the first duration includes:

Generate a time sequence corresponding to T time points detected by each detector in the target road network, and a space sequence corresponding to S road sections in the target road network;

Concatenating the value of the R-type traffic data corresponding to each time point in the time series and the value of the R-type traffic data corresponding to each road segment in the space series to obtain the first tensor.
The method according to claim 1, wherein said utilizing the attention mechanism to fuse the spatial dimension and traffic data dimension in the first tensor to obtain a second tensor, comprising:

Extracting the regional features of the spatial dimension and the traffic data dimension in the first tensor to obtain an intermediate tensor;

The intermediate tensor is input into the attention fusion network to obtain the weight value of each regional feature in the intermediate tensor;

Multiplying the weight value of each region feature with the first tensor to obtain a second tensor.
The method according to claim 1, wherein said use of the attention mechanism fuses the spatial dimension and the traffic data dimension in the first tensor, and after obtaining the second tensor, according to the third tensor, determine the Before the target information of the target road network, the method also includes:

detecting whether there is a missing value in the second tensor;

If there is a vacant value in the second tensor, the vacant value is filled by using a weight-based interpolation method to obtain a third tensor.
The method according to claim 1, wherein, when the target information includes a traffic mode, determining the target information of the target road network according to the third tensor comprises:

Perform tensor decomposition on the third tensor to obtain the first factor matrix corresponding to the time dimension, the second factor matrix corresponding to the space dimension, and the third factor matrix corresponding to the traffic data;

The traffic mode of the target road network is determined according to the first factor matrix, the second factor matrix and the third factor matrix.
The method according to claim 5, wherein said determining the traffic mode of the target road network according to the first factor matrix, the second factor matrix and the third factor matrix comprises:

constructing an objective function that reflects the difference between a first value that is a value of the third tensor and a second value that is based on the first factor matrix , the second factor matrix and the third factor matrix are calculated;

optimizing the first factor matrix, the second factor matrix and the third factor matrix according to the objective function;

The traffic mode of the target road network is determined according to the optimized first factor matrix, second factor matrix and third factor matrix.
The method according to claim 5, wherein the length of the spatial dimension of the third tensor is n road sections, the length of the time dimension is m time points, the length of the traffic data dimension is i type traffic data, n and m is a positive integer, i is an integer greater than 1;

The first factor matrix is an m×Q matrix, the second factor matrix is an n×Q matrix, the third factor matrix is an i×Q matrix, and Q is a positive integer;

The determining the traffic mode of the target road network according to the first factor matrix, the second factor matrix and the third factor matrix includes:

Obtaining Q groups of traffic data sequences according to the first factor matrix, the second factor matrix and the third factor matrix;

Acquiring Q traffic modes corresponding to the Q group of traffic data sequences.
The method according to claim 1, wherein the length of the traffic data dimension of the third tensor is type i traffic data, and i is an integer greater than 1;

In the case where the target information includes traffic laws, the determination of the target information of the target road network according to the third tensor includes at least one of the following:

Obtain the average value of the kth traffic data of the target road network within the target duration;

Obtaining the average value of the kth type of traffic data in the target space of the target road network within the first time period;

Wherein, the time unit of the target duration includes at least one of the following: day; week; year; the spatial unit of the target space includes at least one of the following: lane; road section; road network; the value range of k is 1 to i .
The method according to claim 1, wherein, when the target information includes a future traffic state, determining the target information of the target road network according to the third tensor comprises:

Input the third tensor into the trained first model, and predict the fourth tensor, which is used to reflect the traffic state of the target road network within the second time period, and the second The duration is the next duration of the first duration;

Wherein, the first hidden state when the first model predicts the fourth tensor is determined by the third tensor and the historical hidden state, and the historical hidden state is the hidden state of the first model in the third duration State determination, the third duration is a previous duration of the first duration; the hidden state of the first model is used to determine the output of the first model.
The method according to claim 9, wherein, before inputting the third tensor into the trained first model, before the fourth tensor is predicted, the method further comprises:

Get the sample tensor;

inputting the sample tensor into the untrained first model to obtain a fifth tensor;

determining a loss function corresponding to the first model according to the sample tensor and the fifth tensor;

According to the loss function, the weight value of the first model is adjusted to obtain the trained first model.
A traffic analysis device, comprising:

A generating module, configured to generate a first tensor according to the traffic data detected by each detector in the target road network within the first duration, the length of the spatial dimension of the first tensor being S in the target road network sections, the length of the time dimension is T time points in the first duration, the length of the traffic data dimension is the R type traffic data detected by the detector, S and T are both positive integers, and R is greater than 1 integer;

The fusion module is used to fuse the spatial dimension and the traffic data dimension in the first tensor by using the attention mechanism to obtain the second tensor;

The first determination module is configured to determine the target information of the target road network according to the third tensor;

Wherein, the third tensor is the second tensor or is determined based on the second tensor; the target information includes at least one of the following: traffic mode, traffic law, and future traffic state.
An electronic device, comprising: a transceiver, a memory, a processor, and a program stored on the memory and operable on the processor; wherein, the processor is used to read the program in the memory to implement the following The traffic analysis method according to any one of claims 1 to 10.
A readable storage medium for storing a program, wherein when the program is executed by a processor, the traffic analysis method according to any one of claims 1 to 10 is implemented.