Risk determination method and device of traffic accident, electronic equipment and storage medium

Abstract

The invention discloses a traffic accident risk determination method and device, electronic equipment and a storage medium, and belongs to the technical field of intelligent traffic. In the method, a specified geographic area is divided into a plurality of grids, traffic data of the specified geographic area in past T time periods are acquired, and the risk of a traffic accident is determined based on the traffic data in the T time periods; the method comprises the steps of generating T parts of graph structure data and T parts of grid structure data, enabling any one piece of graph structure data to represent the dependency relationship of traffic accidents on geographic semantics, enabling any one piece of grid structure data to represent the dependency relationship of the traffic accidents on geographic positions, then analyzing the spatial-temporal correlation of the traffic accidents on geographic semantics based on the T parts of graph structure data, and obtaining the spatial-temporal correlation of the traffic accidents on geographic semantics. Based on the T grid structure data, the time-space relevance of the traffic accident in the geographic position is analyzed, and the traffic accident risk value of each grid in the (T + 1) th time period is determined by integrating the analysis results of the T grid structure data and the time-space relevance, so that a traffic accident prediction scheme is provided.

Description

Technical Field

[0001] The present application relates to the field of intelligent transportation technology, and in particular to a method, device, electronic device and storage medium for determining the risk of a traffic accident. Background Art

[0002] With the increasing number of vehicles in cities, traffic accidents are becoming more frequent, so there is a need for traffic accident prediction. Traffic accident prediction is of great significance to urban public safety, emergency response and construction planning, but due to the large amount of traffic accident data and the fact that traffic accidents are affected by many factors, the research is very challenging. Summary of the invention

[0003] The embodiments of the present application provide a method, device, electronic device and storage medium for determining the risk of a traffic accident, so as to provide a traffic accident prediction solution.

[0004] In a first aspect, an embodiment of the present application provides a method for determining the risk of a traffic accident, comprising:

[0005] Obtaining traffic data for a specified geographical area in the past T time periods, where the specified geographical area is divided into a plurality of grids;

[0006] Based on the traffic data in T time periods, T pieces of graph structure data and T pieces of grid structure data are generated, where any graph structure data is used to characterize the geographic semantic dependency of traffic accidents, and any grid structure data is used to characterize the geographic location dependency of traffic accidents;

[0007] Based on T pieces of graph structure data, the temporal and spatial correlation of traffic accidents in geographical semantics is analyzed to obtain the first feature; based on T pieces of grid structure data, the temporal and spatial correlation of traffic accidents in geographical location is analyzed to obtain the second feature;

[0008] Based on the first feature and the second feature, a traffic accident risk value of each grid in the T+1th time period is determined.

[0009] In a second aspect, an embodiment of the present application provides a traffic accident risk determination device, comprising:

[0010] An acquisition module, used to acquire traffic data of a specified geographical area in the past T time periods, wherein the specified geographical area is divided into a plurality of grids;

[0011] A generation module, for generating T pieces of graph structure data and T pieces of grid structure data based on traffic data in T time periods, wherein any graph structure data is used to characterize the geographic semantic dependency of traffic accidents, and any grid structure data is used to characterize the geographic location dependency of traffic accidents;

[0012] An analysis module is used to analyze the spatiotemporal correlation of traffic accidents in geographical semantics based on T pieces of graph structure data to obtain a first feature; and to analyze the spatiotemporal correlation of traffic accidents in geographical locations based on T pieces of grid structure data to obtain a second feature;

[0013] A determination module is used to determine the traffic accident risk value of each grid in the T+1th time period based on the first feature and the second feature.

[0014] In a third aspect, an embodiment of the present application provides an electronic device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein:

[0015] The memory stores a computer program that can be executed by at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute any of the above-mentioned traffic accident risk determination methods.

[0016] In a fourth aspect, an embodiment of the present application provides a storage medium. When a computer program in the storage medium is executed by a processor of an electronic device, the electronic device can execute any of the above-mentioned traffic accident risk determination methods.

[0017] In the embodiment of the present application, the designated geographical area is divided into multiple grids, and the traffic data of the designated geographical area in the past T time periods are obtained. Based on the traffic data in the T time periods, T portions of graph structure data and T portions of grid structure data are generated. Any graph structure data is used to characterize the dependency of traffic accidents in geographical semantics, and any grid structure data is used to characterize the dependency of traffic accidents in geographical locations. Then, based on the T portions of graph structure data, the spatiotemporal correlation of traffic accidents in geographical semantics is analyzed, and based on the T portions of grid structure data, the spatiotemporal correlation of traffic accidents in geographical locations is analyzed. The analysis results of the two are combined to determine the traffic accident risk value of each grid in the T+1th time period, thereby providing a traffic accident prediction scheme. Moreover, the scheme comprehensively considers the spatiotemporal correlation of traffic accidents in geographical semantics and geographical locations to predict traffic accidents, and the prediction accuracy is also relatively high. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0019] Figure 1 A schematic diagram of a process for predicting traffic accident risks at the urban road network level provided in an embodiment of the present application;

[0020] Figure 2A schematic diagram of a data preprocessing process provided in an embodiment of the present application;

[0021] Figure 3 A network structure diagram for mining geographic spatiotemporal dependencies provided in an embodiment of the present application;

[0022] Figure 4 A network structure diagram for mining semantic spatiotemporal dependencies provided in an embodiment of the present application;

[0023] Figure 5 A network structure diagram of a prediction stage provided in an embodiment of the present application;

[0024] Figure 6 A structural diagram of a risk prediction model provided in an embodiment of the present application;

[0025] Figure 7 A flow chart of a method for determining the risk of a traffic accident provided in an embodiment of the present application;

[0026] Figure 8 A flow chart of a method for generating structure data provided in an embodiment of the present application;

[0027] Fig. 9 A flow chart of another method for determining the risk of a traffic accident provided in an embodiment of the present application;

[0028] Fig.10 A schematic diagram of the structure of a traffic accident risk determination device provided in an embodiment of the present application;

[0029] Fig.11 A schematic diagram of the hardware structure of an electronic device for implementing a method for determining the risk of traffic accidents provided in an embodiment of the present application. DETAILED DESCRIPTION

[0030] In order to provide a traffic accident prediction solution, the embodiments of the present application provide a traffic accident risk determination method, device, electronic device and storage medium.

[0031] The preferred embodiments of the present application are described below in conjunction with the drawings in the specification. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application, and are not used to limit the present application. In addition, the embodiments and features in the embodiments of the present application may be combined with each other if there is no conflict.

[0032] In order to facilitate the understanding of this application, the technical terms involved in this application are:

[0033] Convolutional Neural Networks (CNN) are a type of feed-forward neural network that includes convolution calculations and has a deep structure. It is one of the representative algorithms of deep learning.

[0034] An undirected graph, or graph structure, is usually represented by G=(V, E), where V represents a vertex and E represents a set of edges. In the present application, a vertex represents a grid, and two vertices connected by an edge represent that the two grids corresponding to the two vertices have similar geographical semantics, such as both grids are shopping malls, both are large transportation hubs, and both are viaducts with complex terrain.

[0035] Graph Convolutional Network (GCN), the graph structure is relatively complex and generally not neat, which makes traditional neural network operations not well used on the graph structure. In order to apply deep learning methods to graph structures, GNN was proposed. GCN is similar to CNN and has powerful feature learning capabilities. Their essence is also similar.

[0036] Long Short Term Memory (LSTM) is an important time series algorithm and a special recurrent neural network (RNN) that can learn long-term dependencies. It is mainly used to solve the gradient vanishing and gradient exploding problems in the long sequence training process. In short, compared with ordinary RNN, LSTM can perform better in longer sequences.

[0037] The Attention mechanism is a technology that allows the model to focus on important information and fully learn and absorb it. It can be used in any sequence model.

[0038] Predicting traffic accidents in large-scale road network areas such as cities is of great significance to urban public safety, emergency response and construction planning. However, since urban traffic accidents are affected by many factors and have multi-scale dependence on spatiotemporal characteristics, research is very challenging. In addition, existing accident risk prediction methods based on statistical rules or empirical analysis are often limited by artificial experience and limited accident cause analysis, which makes them inapplicable to the simultaneous prediction of accident risks in large-scale road networks.

[0039] Based on this, this application proposes a traffic accident risk prediction method that comprehensively considers the dependencies of grids in geographic space-time and semantic space-time, realizes short-term prediction of traffic accident risks under large-scale road network conditions at the urban level, and supports urban traffic management planning.

[0040] First of all, it should be noted that in the implementation of this application, the traffic data of a specified geographical area, such as a city, in the last T time periods is used to predict the traffic accident risk value of each grid divided into the specified geographical area in the T+1th time period. A time period is such as 1 hour, 2 hours, a day, etc., and the traffic data in a time period may include traffic flow and accident description data, wherein the traffic flow is used to indicate the number of vehicles passing through each intersection in this time period, and the accident description data is used to describe the traffic accident situation on each road in this time period, such as casualties. The traffic flow in several consecutive time periods forms a traffic flow sequence, and the accident description data in the consecutive time periods forms an accident risk sequence.

[0041] The following is an introduction to grid division.

[0042] Assuming that we want to predict the risk value of traffic accidents on each road section in a certain urban area, we can divide the urban area into I×J grids according to the longitude and latitude according to the road distribution of this urban area. Each grid represents a geographical area. The size of the geographical area is 2km×2km. The i-th grid can be represented as g i , i=1,…,I*J, and the number of grids with road sections is defined as N, N≤I×J.

[0043] Subsequently, the traffic accident risk at the city road network level can be predicted, see Figure 1 ,The prediction process mainly includes three stages: data preprocessing stage, feature mining stage, and risk ,prediction stage.

[0044] The following is an introduction to these stages.

[0045] 1. Data preprocessing stage.

[0046] See also Figure 2 , is a schematic diagram of a data preprocessing process, whose input is traffic flow sequence and accident risk sequence, wherein the accident risk sequence can be converted into a risk value sequence, and in order to avoid too many zero values ​​in the risk value sequence causing prediction difficulties, the zero values ​​in the risk value sequence can also be data enhanced. After that, the traffic flow sequence and the risk value sequence after data enhancement are used to generate grid-shaped feature data respectively ( Figure 2 The grid data on the left) and the graphical feature data ( Figure 2 right side of the graph).

[0047] The above process is described in detail below.

[0048] 1) Determine the traffic flow status grid by grid and define the initial risk value.

[0049] Assume that the collection time interval of traffic data is Δt, for example, 1 hour, that is, once every hour, and assume that the traffic data currently collected is in the t-th time period.

[0050] Due to g i There may be multiple intersections, and the traffic flow of at least two of these intersections is collected within this hour. Then, the traffic flow of at least two roads can be aggregated to obtain g in the tth time period i Traffic flow state x i (t), where aggregation processing includes weighted averaging, median, etc.

[0051] At the same time, the number of traffic accidents in each grid can be counted based on the accident description data in the tth time period. Since different accident types have different corresponding risk values, the accident types can be divided according to the number of casualties in traffic accidents. For example, three types of traffic accidents are defined: minor accidents, injury accidents, and fatal accidents, and the corresponding risk values ​​are set to 1, 2, and 5 respectively. Then, g i The accident risk value in the tth time period can be defined as the weighted value of the number of accidents. i If two injury accidents and one minor accident occurred in the tth time period, then g i The risk value in the tth time period is r i (t)=5.

[0052] 2) Perform data enhancement on zero risk values.

[0053] Generally speaking, traffic accidents are low-probability events, which will result in the risk values ​​corresponding to most grids being zero, which is not suitable for effective training of road network-level risk prediction models. To this end, the present application embodiment proposes a data enhancement strategy based on prior knowledge to improve the discriminability of zero risk values.

[0054] Specifically, if g i If the risk value corresponding to the t-th time period is zero, it can be converted to a negative value. The conversion can include the following two steps:

[0055] The first step is to convert the zero value into the accident risk indicator ε i (t):

[0056]

[0057] Where T′ represents the number of time periods, and T′>T, r i(j) Table g i The risk value of a traffic accident occurring in the jth time period; r k (j) represents the risk value of a traffic accident occurring in grid k during the jth time period, and N is the number of grids with road sections in the specified geographical area;

[0058] The second step is to convert the index value into statistical accident intensity:

[0059] R i (t) = b1log2ε i (t)+b2,

[0060] Among them, b1 and b2 are to maintain R i (t) A coefficient that is symmetric between the absolute value range and the true risk value range, determined in advance.

[0061] In this way, the logarithmic transformation between 0 and 1 can make the zero-item risk value negative, reflecting that the accident risk of the corresponding grid is low, and lower than all non-negative risk values, which just conforms to the actual situation. Moreover, the smaller the negative value, the lower the accident risk of the corresponding grid, which maintains the actual accident risk level, so that the converted negative risk value is discriminative and suitable for training the network.

[0062] 3) Construct graph structure data.

[0063] Considering that non-adjacent grids may have spatiotemporal similarity due to geographical semantic similarities such as road features and area types (all shopping malls, all transportation hubs), the spatiotemporal similarity of risks between grids can be described from the semantic aspect. The risk undirected graph (or risk similarity graph) G = (V, E, A) is defined to represent the risk similarity pattern between grids, where V is a vertex and represents a grid. The data of the vertex includes the traffic flow and risk value of the grid. E is an edge and represents the similarity relationship between the two connected grids. Represents the similarity matrix.

[0064] To calculate the similarity matrix A, we can first calculate any vertex v i and vertex v j The similarity between them is defined by Jensen-Shannon divergence. For the risk similarity matrix, (v i ,v j ) is defined as follows:

[0065] Sim(v i ,v j )=1-JS(R i ,R j ),

[0066]

[0067] in, Represents the risk distribution of vertices i and j in T′ time periods.

[0068] After obtaining the pairwise similarities of all vertices, the top-1 most similar vertex can be selected as the first-order neighbor for each vertex. Then, the similarity matrix A can be constructed using the similarity between each vertex and its first-order neighbor.

[0069] At this point, the risk undirected graph G = (V, E, A) has been constructed.

[0070] Subsequently, after obtaining the traffic flow and risk value of each grid in the tth time period, the traffic flow and risk value are filled into the constructed risk undirected graph grid by grid, and the graph structure data of the tth time period can be obtained.

[0071] 4) Construct grid structure data.

[0072] After obtaining the traffic flow and risk value of each grid in the tth time period, the traffic flow and risk value of the grids are spliced ​​according to the grid position number to obtain the grid structure data of the tth time period.

[0073] Finally, the traffic flow data x of I×J grids in the tth time period i (t) and the accident risk value r i (t) After data preprocessing, the structural data in the tth time period can be obtained: grid structure data and graph structured data As input for subsequent feature modeling.

[0074] 2. Feature mining stage.

[0075] The main task is to mine the geographic spatiotemporal dependencies and semantic spatiotemporal dependencies of traffic accidents in different regions, see Figure 3 , for geographical spatiotemporal dependencies, convolutional neural networks and attention recurrent neural networks are used to mine, see Figure 4 ,For semantic spatiotemporal dependencies, graph convolutional neural networks and attention recurrent neural networks are used to mine.

[0076] They are introduced below one by one.

[0077] 1) Mining geographic spatiotemporal dependencies.

[0078] When processing the grid structure data X in time period t t When the convolution operation is used to model the spatial features of the local area, for the input X t , the output after the convolution operation is calculated as:

[0079]

[0080] Among them, * represents the convolution operation, and denote the weight and bias parameters of the convolutional layer, respectively, f(·) denotes the ReLU activation function, d c Indicates the number of convolution kernels.

[0081] In addition to the spatial correlation between adjacent grids, traffic accidents and their associated traffic indicators such as traffic flow usually have short-term and long-term correlations in the time dimension. In order to mine these correlations, LSTM can be used to model the short-term and long-term correlations between traffic accidents and their associated traffic indicators in time series.

[0082] For example, for g i , the LSTM calculation process is simplified to:

[0083]

[0084] in, represents the convolutional mining g at the tth time period i The spatial characteristics of d h Represents the number of LSTM hidden units, each LSTM unit contains a storage unit C t , for storage unit C t The access to will be controlled by three gates, namely the forget gate f t , input gate q t and output gate o t Specifically, the update formula of the LSTM unit at time t is expressed as:

[0085]

[0086]

[0087]

[0088]

[0089]

[0090]

[0091] Where σ represents the sigmoid activation function, {W i,q ,W i,f ,W i,o ,W i,g} and {U i,q,U i,f ,U i,o ,U i,g} represents the hidden layer weight, {b i,g ,b i,f ,b i,o ,b i,g} represents the bias vector of the corresponding neuron.

[0092] In actual scenarios, traffic data from different time periods have different impacts on traffic accidents in the target time period, and they change over time. In order to highlight the different importance of traffic data from different time periods to the risk of traffic accidents in the target time period, a temporal self-attention mechanism can be used to dynamically capture the impact of historical information on the risk of traffic accidents in the target time period.

[0093] For example, for g i , the input of the self-attention layer The corresponding association weight is calculated as:

[0094]

[0095] in, represents the LSTM network output within the overall observation period, ψ(·) represents the softmax activation function, represents the ReLU activation function, and represents the weight parameter to be learned, b a Represents the bias vector to be learned.

[0096] The output after attention weighting is:

[0097]

[0098] Then, for all I×J grids, after geospatial dependency mining, the output geospatial temporal correlation feature (i.e., the first feature) is

[0099] 2) Semantic spatiotemporal dependency mining.

[0100] When processing the graph structure data S in time period t t When , first calculate the graph Laplacian matrix Where D represents the degree matrix of the risk undirected graph, I N represents the unit matrix and A is the similarity matrix.

[0101] Next, two layers of GCN are used to learn graph network feature dependencies on the risk similarity graph G. The graph convolution operation is as follows:

[0102]

[0103]

[0104] Among them, l represents the number of network layers, W l ,b l represents weight and bias parameters, L represents the Laplacian matrix, S t Represents the graph features in the tth time period. Through the above operations, we can get the semantic graph embedding

[0105] Convolution hides features in output graph After that, the attention recurrent neural network can be used to model the long-term and short-term temporal features of semantics. i , its graph features Input to the LSTM unit, the calculation process is expressed as:

[0106]

[0107] in, d s Represents the number of LSTM hidden units. Each LSTM unit contains a storage unit M t , for storage unit M t The access to will be controlled by three gates, namely the forget gate f t , input gate q t and output gate o t At each time t, the update formula of the LSTM unit is expressed as,

[0108]

[0109]

[0110]

[0111]

[0112]

[0113]

[0114] Where σ represents the sigmoid activation function, {W i,q ,W i,f ,W i,o ,W i,g} and {U i,q ,U i,f ,U i,o ,U i,g} represents the hidden layer weight, {b i,g ,bi,f ,b i,o ,b i,g} represents the bias vector of the corresponding neuron.

[0115] Similarly, considering that the traffic data of different time periods have different impacts on the traffic accidents in the target time period, and the impacts will change over time. In order to highlight the different importance of different time periods to the traffic accident risk in the target time period, the temporal self-attention mechanism is used to dynamically capture the impact of historical information on the traffic accident risk in the target time period. i , the input of the self-attention layer The corresponding association weight is calculated as:

[0116]

[0117] in, represents the LSTM network output within the overall observation period, ψ(·) represents the softmax activation function, represents the ReLU activation function, and represents the weight parameter to be learned, b s represents the bias vector to be learned. Then, the output after attention weighting is calculated as

[0118] Then, for a graph network containing N vertices, the output after semantic space dependency mining is expressed as By using the matching mapping relationship between grid numbers and graph nodes, the semantic spatiotemporal correlation features (i.e., the second feature) after spatiotemporal semantic modeling can be finally obtained.

[0119] 3. Risk prediction stage.

[0120] See also Figure 5 In the risk prediction stage, external environmental features such as point of interest (POI), weather features, time features, etc. can also be considered. After these features pass through the fully connected layer FC, the external environmental feature vector z of the T+1 time period is obtained. t+1 , d e It represents the dimension after the external environment features are embedded. Then, the external environment features, geographic spatiotemporal association features and semantic spatiotemporal association features are combined to predict the traffic accident risk of each grid in the target time period (such as the T+1 time period). That is, the traffic accident risk value of each grid in the T+1 time period is:

[0121]

[0122] Among them, * represents a 1×1 convolution kernel, W1, W2, and W3 represent the weight parameters of the convolution kernel, and FC(·) represents a fully connected layer.

[0123] Considering that traffic accidents show multi-scale dependencies in both geographic space-time and semantic space-time, the embodiment of this application proposes a hybrid deep neural network model, which uses CNN and Attention-LSTM to model geographic space-time dependencies, and uses GCN and Attention-LSTM to model semantic space-time dependencies, and finally integrates with external environment embedded features to achieve short-term prediction of road network-level traffic accident risks. In addition, a new data sample enhancement strategy is proposed to address the zero-inflation problem faced by urban traffic accident risk prediction. It does not rely on prior hyperparameters, but directly uses the statistical prior characteristics of historical data to enhance data samples, supporting subsequent deep network training and model learning.

[0124] The following is an example of predicting the traffic accident risk in city C to introduce the solution of the embodiment of the present application.

[0125] Assume that the traffic data of city C in the past month is used to construct sample data, and the traffic data of city C in the past 48 consecutive hours is used to predict the accident risk of city C in the next 49 hours. Assuming that the data collection cycle is 1 hour, that is, one time period is one hour, then T′=720, T=48.

[0126] The first step is to divide the grid.

[0127] According to the road distribution of city C, the area of ​​city C can be divided into I×J grids according to longitude and latitude. The i-th grid can be expressed as g i , i=1,…,I*J, and the number of grids with road sections is defined as N, N≤I×J.

[0128] In the second step, a risk undirected graph is constructed based on the semantic similarity of the grids.

[0129] Define a risk undirected graph G = (V, E, A), where V is a vertex, representing a grid. The data of the vertex may include the traffic flow and risk value of the grid. E is an edge, representing the similarity relationship between the two connected grids. Represents the similarity matrix.

[0130] To calculate the similarity matrix A, we can first calculate any vertex v i and vertex v j The similarity between them is defined by Jensen-Shannon divergence. For the risk similarity matrix, (v i ,v j ) is defined as follows:

[0131] Sim(v i ,v j )=1-JS(R i ,R j ),

[0132]

[0133] in, Represents the risk distribution of vertices i and j in T′ time periods.

[0134] After obtaining the pairwise similarities of all vertices, the top-1 most similar vertex can be selected as the first-order neighbor for each vertex. Then, the similarity matrix A can be constructed using the similarity between each vertex and its first-order neighbor. At this point, the risk undirected graph G = (V, E, A) is constructed.

[0135] The third step is to preprocess the sample data.

[0136] Get the traffic data of city C within any continuous 48 hours in the past month as a sample data, that is, there are traffic data within 48 time periods, which can be seen in Figure 2 ,According to the order of collection time from morning to night, the traffic flow in these 48 time periods forms a traffic flow sequence, and the accident description data in these 48 time periods forms an accident risk sequence.

[0137] Afterwards, the traffic flow sequence corresponding to each grid can be determined based on the traffic flow sequence, and the accident risk sequence corresponding to each grid can be determined based on the accident risk sequence. According to the rule that the severity of the accident is positively correlated with the risk value, the accident risk sequence corresponding to this grid can be converted into a risk value sequence. To avoid the zero inflation problem, the zero value in the risk value sequence can also be converted into a negative number.

[0138] For example, the zero values ​​in the risk value series are converted to negative numbers according to the following formula:

[0139]

[0140] R i (t) = b1log2ε i (t)+b2,

[0141] Among them, r i (j) represents the risk value of a traffic accident occurring in grid i during the jth time period, r k (j) represents the risk value of a traffic accident occurring in grid k during the jth time period, N is the number of grids with road sections in the specified geographical area, and R i(t) is the risk value of grid i converted into a negative number in the tth time period, b1 and b2 are pre-selected hyperparameters, 1≤j≤T.

[0142] After obtaining the traffic flow sequence and the converted risk value sequence corresponding to each grid, the traffic flow and risk value corresponding to each grid in these sequences in each time period are used to fill in the pre-constructed risk undirected graph to obtain 1 graph structure data. There are 48 time periods in total, so 48 graph structure data can be obtained; according to the grid position number, the traffic flow and risk value corresponding to each grid in these sequences in each time period are spliced ​​to obtain 1 grid structure data. There are 48 time periods in total, so 48 grid structure data can be obtained.

[0143] At this point, the preprocessing of a sample data is completed. In addition, it should be noted that a sample data also has a risk labeling value, which can be pre-determined by a technician based on experience or experiments.

[0144] Multiple data samples are obtained from the traffic data of city C in the past month and preprocessed according to the above process to prepare for subsequent model training.

[0145] The fourth step is to use samples to train the risk prediction model.

[0146] See also Figure 6 In the feature mining stage, CNN is used to extract features from 48 grid structure data one by one, LSTM is used to perform temporal correlation analysis on the 48 feature extraction results, and the Attention mechanism is used to adjust the importance of different time periods in the analysis results to obtain geographic spatiotemporal correlation features. GNN is used to extract features from 48 graph structure data one by one, LSTM is used to perform temporal correlation analysis on the 48 feature extraction results, and the Attention mechanism is used to adjust the importance of different time periods in the analysis results to obtain semantic spatiotemporal correlation features.

[0147] Continue to see Figure 6 In the prediction stage, the POI features, weather features, and time features of city C in the next hour are obtained. After these features are FC, the external environment feature vector of city C in the next hour can be obtained. The external environment feature vector, geographic spatiotemporal association features, and semantic spatiotemporal association features are input into the fusion layer for fusion. After fusion, the FC layer is passed to obtain the risk prediction value of traffic accidents in each grid divided by city C in the next hour. The risk prediction value is compared with the risk labeling value. According to the comparison results, the network parameters in the feature mining layer and the prediction layer are adjusted until the training end conditions are met, such as when the model accuracy reaches the set accuracy or the number of training times reaches the set number, the training can be ended to obtain a risk prediction model.

[0148] The fifth step is to preprocess the prediction data.

[0149] Here, the prediction data refers to the traffic data within the nearest 48 hours to the current time period. In the actual process, the preprocessing process of the prediction data is similar to the preprocessing process of the sample data in the third step, which will not be repeated here.

[0150] Step 6: Use the predicted data and the trained risk prediction model to predict the traffic accident risk for city C.

[0151] The preprocessed prediction data is input into the trained risk prediction model, and the output is the risk value of traffic accidents occurring in each grid divided by city C in the current time period.

[0152] The embodiment of the present application provides a risk prediction method for large-scale urban road networks, and proposes a short-term accident risk prediction model based on a deep neural network for the problem of traffic accident risk prediction for large-scale urban road networks. The method specifically proposes a road network-level risk prediction model that can jointly characterize geographic spatial dependency and semantic spatial dependency. In this model, a priori knowledge strategy is first used to enhance traffic accident risk data, and zero-inflated data samples are converted into differentiated data samples suitable for neural network learning. Then, the impact of geographic spatial dependency and semantic spatial dependency on traffic accident risk is characterized at the feature mining layer, and convolutional neural networks and attention recurrent neural networks are used to model geographic spatiotemporal dependency features. At the same time, graph convolutional networks and attention recurrent neural networks are used to model semantic spatiotemporal dependency features. Finally, the risk prediction layer combines external environment embedded features including POI, weather, time, etc. to output the traffic accident risk value for the next time period. The factors considered are relatively comprehensive, so the accuracy of risk prediction is relatively high.

[0153] Figure 7 A flow chart of a method for determining the risk of a traffic accident provided in an embodiment of the present application, the method comprising the following steps.

[0154] In step 701, traffic data of a specified geographical area in the past T time periods is obtained, and the specified geographical area is divided into a plurality of grids.

[0155] For example, traffic data in the T time periods closest to the current time period can be obtained. In this way, using the latest traffic data for prediction can help improve prediction accuracy.

[0156] In step 702, based on the traffic data in T time periods, T pieces of graph structure data and T pieces of grid structure data are generated, where any graph structure data is used to characterize the geographic semantic dependencies of traffic accidents, and any grid structure data is used to characterize the geographic location dependencies of traffic accidents.

[0157] In some embodiments, the traffic data in each time period may include traffic flow and accident description data, which can be Figure 8 The process shown generates structure data, which includes the following steps:

[0158] In step 7021, traffic flow and accident description data corresponding to each grid are selected from the traffic data in the tth time period.

[0159] Here, t is an integer from 1 to T.

[0160] In step 7022, according to the rule that the severity of the accident is positively correlated with the risk value, the accident description data is converted into the risk value corresponding to the grid.

[0161] For example, three types of traffic accidents are defined: minor accidents, injury accidents, and fatal accidents, and the corresponding risk values ​​are set to 1, 2, and 5. Then, the accident risk value of a grid in the tth time period can be defined as the weighted value of the number of accidents. Assuming that a grid has 2 injury accidents and 1 minor accident in the tth time period, the risk value of this grid in the tth time period is 5.

[0162] In step 7023, if the risk value corresponding to the grid is zero, the zero value is converted into a negative number.

[0163] For example, convert zero values ​​to negative numbers according to the following formula:

[0164]

[0165] R i (t) = b1log2ε i (t)+b2,

[0166] Where T′ represents the number of time periods, and T′>T, r i (j) represents the risk value of a traffic accident occurring in grid i during the jth time period, r k (j) represents the risk value of a traffic accident occurring in grid k during the jth time period, N is the number of grids with road sections in the specified geographical area, and R i (t) is the risk value of grid i converted into a negative number in the tth time period, b1 and b2 are pre-selected hyperparameters, 1≤j≤T.

[0167] In step 7024, a grid structure data is generated based on the traffic flow and risk value corresponding to each grid.

[0168] For example, by concatenating the traffic flow and risk value of each grid according to the grid position number, the grid structure data of the tth time period can be obtained.

[0169] In step 7025, a graph structure data is generated based on the traffic flow and risk value corresponding to each grid, and the traffic flow and risk value corresponding to the grids that have a connection relationship in the undirected graph. The undirected graph is pre-constructed based on the geographic semantic similarity between the grids.

[0170] For example, by filling the traffic flow and risk value of each grid in the tth time period into the pre-constructed risk undirected graph grid by grid, the graph structure data of the tth time period can be obtained. The construction process of the risk undirected graph can be found in the previous introduction and will not be repeated here.

[0171] It should be noted that there is no strict sequence relationship between steps 7024 and 7025.

[0172] In step 703, based on T portions of graph structure data, the temporal and spatial correlation of traffic accidents in geographic semantics is analyzed to obtain a first feature; based on T portions of grid structure data, the temporal and spatial correlation of traffic accidents in geographic location is analyzed to obtain a second feature.

[0173] In some embodiments, GCN can be used to extract features from T pieces of graph structure data, LSTM can be used to perform time correlation analysis on T feature extraction results, and the Attention mechanism can be used to adjust the importance of different time periods in the analysis results to obtain the first feature.

[0174] In some embodiments, CNN can be used to extract features from T grid structure data, LSTM can be used to perform time correlation analysis on T feature extraction results, and the Attention mechanism can be used to adjust the importance of different time periods in the analysis results to obtain a second feature.

[0175] In step 704, based on the first feature and the second feature, the traffic accident risk value of each grid in the T+1th time period is determined.

[0176] For example, the first feature and the second feature are input into the fusion layer to obtain the fusion feature, and the fusion feature is input into the FC layer to obtain the traffic accident risk value of each grid in the T+1th time period.

[0177] Fig. 9 A flowchart of another method for determining the risk of a traffic accident provided in an embodiment of the present application, the method comprising the following steps.

[0178] In step 901, traffic data of a specified geographical area in the past T time periods is obtained, and the specified geographical area is divided into a plurality of grids.

[0179] In step 902, based on the traffic data in T time periods, T pieces of graph structure data and T pieces of grid structure data are generated, where any graph structure data is used to characterize the geographic semantic dependencies of traffic accidents, and any grid structure data is used to characterize the geographic location dependencies of traffic accidents.

[0180] The implementation of this step can refer to step 702 and will not be described in detail here.

[0181] In step 903, based on T portions of graph structure data, the temporal and spatial correlation of traffic accidents in geographic semantics is analyzed to obtain a first feature; based on T portions of grid structure data, the temporal and spatial correlation of traffic accidents in geographic location is analyzed to obtain a second feature.

[0182] The implementation of this step can refer to step 703, which will not be described in detail here.

[0183] In step 904, at least one accident risk characterization data of a specified geographical area in a T+1th time period is obtained.

[0184] Among them, accident risk characterization data such as POI, weather, date, etc.

[0185] In step 905, the embedded features of various accident risk characterization data are fused to obtain a third feature.

[0186] In step 906, the first feature, the second feature and the third feature are fused to obtain a fused feature.

[0187] In step 907, the fusion features are analyzed to obtain the traffic accident risk value of each grid in the T+1th time period.

[0188] In this way, the accident risk characterization data in the T+1th time period is additionally considered, and the information considered is more comprehensive, which is conducive to improving the prediction accuracy.

[0189] It should be noted that there is no strict sequence relationship between steps 904-905 and steps 901-903.

[0190] Based on the same technical concept, an embodiment of the present application also provides a risk determination device for traffic accidents. The principle of solving the problem by the risk determination device for traffic accidents is similar to the risk determination method for traffic accidents mentioned above. Therefore, the implementation of the risk determination device for traffic accidents can refer to the implementation of the risk determination method for traffic accidents, and the repeated parts will not be repeated.

[0191] Fig.10 A structural schematic diagram of a traffic accident risk determination device provided in an embodiment of the present application includes an acquisition module 1001, a generation module 1002, an analysis module 1003, and a determination module 1004.

[0192] An acquisition module 1001 is used to acquire traffic data of a specified geographical area in the past T time periods, where the specified geographical area is divided into a plurality of grids;

[0193] A generating module 1002 is used to generate T pieces of graph structure data and T pieces of grid structure data based on traffic data in T time periods, wherein any graph structure data is used to characterize the geographical semantic dependency of traffic accidents, and any grid structure data is used to characterize the geographical dependency of traffic accidents;

[0194] The analysis module 1003 is used to analyze the temporal and spatial correlation of traffic accidents in geographical semantics based on T pieces of graph structure data to obtain a first feature; and to analyze the temporal and spatial correlation of traffic accidents in geographical locations based on T pieces of grid structure data to obtain a second feature;

[0195] The determination module 1004 is used to determine the traffic accident risk value of each grid in the T+1th time period based on the first feature and the second feature.

[0196] In some embodiments, the traffic data in each time period includes traffic flow and accident description data, and the generation module 1002 is specifically used to:

[0197] Select the traffic flow and accident description data corresponding to each grid from the traffic data in the t-th time period, 1≤t≤T;

[0198] According to the rule that the severity of the accident is positively correlated with the risk value, the accident description data is converted into the risk value corresponding to the grid;

[0199] Generate a grid structure data based on the traffic flow and risk value corresponding to each grid;

[0200] A graph structure data is generated based on the traffic flow and risk value corresponding to each grid and the traffic flow and risk value corresponding to the grids having a connection relationship in an undirected graph, wherein the undirected graph is pre-constructed according to the geographic semantic similarity between the grids.

[0201] In some embodiments, the generating module 1002 is further configured to:

[0202] After converting the accident description data into the risk value corresponding to the grid, if the risk value corresponding to the grid is zero, the zero value is converted into a negative number.

[0203] In some embodiments, the generating module 1002 is specifically used to:

[0204] The zero value is converted to a negative number according to the following formula:

[0205]

[0206] R i (t) = b1log2ε i (t)+b2,

[0207] Where T′ represents the number of time periods, and T′>T, r i (j) represents the risk value of a traffic accident occurring in grid i during the jth time period, r k (j) represents the risk value of a traffic accident occurring in grid k in the jth time period, N is the number of grids with road sections contained in the specified geographical area, R i (t) is the risk value of grid i converted into a negative number in the tth time period, b1 and b2 are pre-selected hyperparameters, 1≤j≤T.

[0208] In some embodiments, the analysis module 1003 is specifically used to:

[0209] Use graph convolutional network GCN to extract features from T pieces of graph structure data in turn;

[0210] Use the long short-term memory network LSTM to perform temporal correlation analysis on T feature extraction results;

[0211] The attention mechanism is used to adjust the importance of different time periods in the analysis results to obtain the first feature.

[0212] In some embodiments, the analysis module 1003 is specifically used to:

[0213] The convolutional neural network (CNN) is used to extract features from T grid structure data in turn;

[0214] Use the long short-term memory network LSTM to perform temporal correlation analysis on T feature extraction results;

[0215] The attention mechanism is used to adjust the importance of different time periods in the analysis results to obtain the second feature.

[0216] In some embodiments, it also includes:

[0217] The processing module 1005 is used to obtain at least one accident risk characterization data of the specified geographical area in the T+1th time period; and fuse the embedded features of various accident risk characterization data to obtain a third feature;

[0218] The determination module 1004 is further used to fuse the first feature, the second feature and the third feature to obtain a fused feature; and analyze the fused feature to obtain a traffic accident risk value of each grid in the T+1th time period.

[0219] The division of modules in the embodiments of the present application is schematic and is only a logical function division. There may be other division methods in actual implementation. In addition, the functional modules in the embodiments of the present application may be integrated in a processor, or may exist physically separately, or two or more modules may be integrated in one module. The coupling between the modules may be achieved through some interfaces, which are usually electrical communication interfaces, but it is not excluded that they may be mechanical interfaces or other forms of interfaces. Therefore, the modules described as separate components may or may not be physically separated, and may be located in one place or distributed to different locations of the same or different devices. The above-mentioned integrated modules may be implemented in the form of hardware or in the form of software functional modules.

[0220] After introducing the method and device for determining the risk of a traffic accident according to an exemplary embodiment of the present application, next, an electronic device according to another exemplary embodiment of the present application is introduced.

[0221] Refer to the following Fig.11 The electronic device 130 implemented according to this embodiment of the present application is described. Fig.11 The electronic device 130 shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0222] like Fig.11 As shown, the electronic device 130 is in the form of a general electronic device. The components of the electronic device 130 may include but are not limited to: the at least one processor 131, the at least one memory 132, and a bus 133 connecting different system components (including the memory 132 and the processor 131).

[0223] Bus 133 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, a processor, or a local bus using any of a variety of bus architectures.

[0224] The memory 132 may include a readable medium in the form of a volatile memory, such as a random access memory (RAM) 1321 and / or a cache memory 1322 , and may further include a read-only memory (ROM) 1323 .

[0225] The memory 132 may also include a program / utility 1325 having a set (at least one) of program modules 1324, such program modules 1324 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0226] The electronic device 130 may also communicate with one or more external devices 134 (e.g., keyboards, pointing devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 130, and / or communicate with any device that enables the electronic device 130 to communicate with one or more other electronic devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 135. Furthermore, the electronic device 130 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 136. As shown, the network adapter 136 communicates with other modules for the electronic device 130 via a bus 133. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 130, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0227] In an exemplary embodiment, a storage medium is also provided, and when a computer program in the storage medium is executed by a processor of an electronic device, the electronic device can perform the above-mentioned traffic accident risk determination method. Optionally, the storage medium can be a non-temporary computer-readable storage medium, for example, the non-temporary computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0228] In an exemplary embodiment, the electronic device of the present application may include at least one processor and a memory communicatively connected to the at least one processor, wherein the memory stores a computer program executable by the at least one processor, and when the computer program is executed by the at least one processor, the at least one processor may execute the steps of any traffic accident risk determination method provided in the embodiments of the present application.

[0229] In an exemplary embodiment, a computer program product is also provided. When the computer program product is executed by an electronic device, the electronic device can implement any exemplary method provided in the present application.

[0230] Furthermore, the computer program product may use any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a RAM, a ROM, an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a compact disk read only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0231] The program product for determining the risk of traffic accidents in the embodiment of the present application may adopt a CD-ROM and include program code, and may be run on a computing device. However, the program product of the present application is not limited thereto, and in this document, a readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, apparatus, or device.

[0232] The readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, wherein the readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The readable signal medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0233] The program code contained on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, radio frequency (RF), etc., or any suitable combination of the foregoing.

[0234] Program code for performing the operations of the present application may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device via any type of network such as a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0235] It should be noted that, although several units or subunits of the device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided into multiple units to be embodied.

[0236] In addition, although the operations of the method of the present application are described in a specific order in the drawings, this does not require or imply that the operations must be performed in this specific order, or that all the operations shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps.

[0237] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0238] The present application is described with reference to the flowcharts and / or block diagrams of the methods, apparatus (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0239] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0240] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0241] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0242] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application also includes these modifications and variations.

Claims

1. A method for determining the risk of a traffic accident, characterized in that: include: Obtaining traffic data of a specified geographical area in the past T time periods, where the specified geographical area is divided into a plurality of grids; Based on the traffic data in T time periods, T pieces of graph structure data and T pieces of grid structure data are generated, where any graph structure data is used to characterize the geographic semantic dependency of traffic accidents, and any grid structure data is used to characterize the geographic location dependency of traffic accidents; Based on T-part graph structure data, the temporal and spatial correlation of traffic accidents in geographical semantics is analyzed to obtain the first feature; Based on T pieces of grid structure data, the temporal and spatial correlation of traffic accidents in geographical locations is analyzed to obtain the second feature; Based on the first feature and the second feature, a traffic accident risk value of each grid in the T+1th time period is determined.

2. The method according to claim 1, characterized in that The traffic data in each time period includes traffic flow and accident description data. Based on the traffic data in T time periods, T pieces of graph structure data and T pieces of grid structure data are generated, including: Select the traffic flow and accident description data corresponding to each grid from the traffic data in the t-th time period, 1≤t≤T; According to the rule that the severity of the accident is positively correlated with the risk value, the accident description data is converted into the risk value corresponding to the grid; Generate a grid structure data based on the traffic flow and risk value corresponding to each grid; A graph structure data is generated based on the traffic flow and risk value corresponding to each grid and the traffic flow and risk value corresponding to the grids having a connection relationship in an undirected graph, wherein the undirected graph is pre-constructed according to the geographic semantic similarity between the grids.

3. The method according to claim 2, characterized in that After converting the accident description data into the risk value corresponding to the grid, the method further includes: If the risk value corresponding to the grid is zero, the zero value is converted into a negative number.

4. The method according to claim 3, characterized in that The zero value is converted to a negative number according to the following formula: R i (t)=b1log2ε i (t)+b2, Among them, T ′ represents the number of time periods, and T ′ >T,r i (j) represents the risk value of a traffic accident occurring in grid i during the jth time period, r k (j) represents the risk value of a traffic accident occurring in grid k in the jth time period, N is the number of grids with road sections contained in the specified geographical area, R i (t) is the risk value of grid i converted into a negative number in the tth time period, b1 and b2 are pre-selected hyperparameters, 1≤j≤T.

5. The method according to claim 1, characterized in that Based on T-part graph structure data, the temporal and spatial correlation of traffic accidents in geographical semantics is analyzed to obtain the first feature, including: Use graph convolutional network GCN to extract features from T pieces of graph structure data in turn; Use the long short-term memory network LSTM to perform temporal correlation analysis on T feature extraction results; The attention mechanism is used to adjust the importance of different time periods in the analysis results to obtain the first feature.

6. The method according to claim 1, characterized in that Based on T pieces of grid structure data, the temporal and spatial correlation of traffic accidents in geographical locations is analyzed to obtain the second feature, including: The convolutional neural network (CNN) is used to extract features from T grid structure data in turn; Use the long short-term memory network LSTM to perform temporal correlation analysis on T feature extraction results; The attention mechanism is used to adjust the importance of different time periods in the analysis results to obtain the second feature.

7. The method according to claim 1, characterized in that Also includes: Obtaining at least one type of accident risk characterization data for the designated geographical area in the T+1th time period; The embedded features of various accident risk characterization data are fused to obtain the third feature; Determining the traffic accident risk value of each grid in the T+1th time period based on the first feature and the second feature includes: fusing the first feature, the second feature and the third feature to obtain a fused feature; The fusion features are analyzed to obtain the traffic accident risk value of each grid in the T+1th time period.

8. A traffic accident risk determination device, characterized in that: include: An acquisition module, used to acquire traffic data of a specified geographical area in the past T time periods, wherein the specified geographical area is divided into a plurality of grids; A generation module, for generating T pieces of graph structure data and T pieces of grid structure data based on traffic data in T time periods, wherein any graph structure data is used to characterize the geographic semantic dependency of traffic accidents, and any grid structure data is used to characterize the geographic location dependency of traffic accidents; An analysis module is used to analyze the spatiotemporal correlation of traffic accidents in geographic semantics based on T pieces of graph structure data to obtain a first feature; Based on T pieces of grid structure data, the temporal and spatial correlation of traffic accidents in geographical locations is analyzed to obtain the second feature; A determination module is used to determine the traffic accident risk value of each grid in the T+1th time period based on the first feature and the second feature.

9. An electronic device, characterized in that: include: at least one processor, and a memory communicatively connected to the at least one processor, wherein: The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can perform the method according to any one of claims 1 to 7.

10. A storage medium, characterized in that: When the computer program in the storage medium is executed by a processor of an electronic device, the electronic device can execute the method as claimed in any one of claims 1 to 7.

Images