Traffic accident risk prediction method, electronic device, storage medium, and program product

Abstract

The present disclosure provides a traffic accident risk prediction method, an electronic device, a storage medium and a program product. The method of the present disclosure comprises: constructing a traffic accident knowledge graph based on historical traffic accident data, and determining the embedding vectors of each entity; determining the accident statistical characteristics corresponding to different times and different regions based on the historical traffic accident data; matching the region with the region entity, obtaining the embedding vectors of the matched region entity as the semantic embedding vectors of the region; fusing the accident statistical characteristics and the semantic embedding vectors to obtain the joint feature vectors; constructing a time series tensor based on the joint feature vectors; constructing a weighted neighbor space graph based on the similarity between the semantic embedding vectors corresponding to each region in the time series tensor; and performing traffic accident risk prediction based on the time series tensor and the weighted neighbor space graph to obtain a traffic accident risk prediction result. The present disclosure can realize accurate prediction of traffic accident risk.

Description

Technical Field

[0001] This disclosure relates to the field of traffic safety technology, specifically to a method for predicting traffic accident risks, electronic equipment, storage media, and program products. Background Technology

[0002] Traffic accidents, as a major public safety issue causing personal injury and property damage, exhibit significant spatiotemporal dependence and the coupling of multiple influencing factors. With accelerated urbanization and a surge in motor vehicle ownership, accurately predicting accident probabilities and severity is crucial for improving road traffic safety management and achieving risk warning and resource optimization in intelligent transportation systems. Existing technologies primarily employ traditional statistical, machine learning, and deep learning spatiotemporal models for prediction. While these models offer some interpretability and reveal macroscopic spatiotemporal correlations with small-scale data, they often rely on the assumption of linear stationarity, making it difficult to effectively capture high-dimensional nonlinear characteristics and complex interactions between multiple variables. Furthermore, they cannot accurately characterize the fine-grained spatiotemporal features of the dynamic evolution of traffic systems. In addition, existing methods suffer from high computational complexity, strong dependence on large-scale data and hardware resources, and a lack of dynamic fusion mechanisms for external heterogeneous information, resulting in insufficient model interpretability and real-time update capabilities. In summary, existing technologies struggle to adapt to real-time changing spatiotemporal environments and have limitations in handling large-scale, multi-source, heterogeneous data and complex variables, hindering further improvements in the accuracy of traffic accident risk prediction. Summary of the Invention

[0003] This disclosure provides a method for predicting traffic accident risks, an electronic device, a storage medium, and a program product.

[0004] According to one aspect of this disclosure, a method for predicting traffic accident risks is provided, comprising: constructing a traffic accident knowledge graph based on historical traffic accident data, and determining the embedding vector of each entity in the traffic accident knowledge graph; determining accident statistical features corresponding to different times and regions based on historical traffic accident data; matching the regions with regional entities in the traffic accident knowledge graph, and obtaining the embedding vector of the matched regional entities as the semantic embedding vector of the region; fusing the accident statistical features with the semantic embedding vector of the corresponding region to obtain a joint feature vector of each region at different times; constructing a temporal tensor containing time and regional dimensions based on the joint feature vector; constructing a weighted nearest neighbor spatial graph based on the similarity between the semantic embedding vectors corresponding to each region in the temporal tensor; and performing traffic accident risk prediction based on the temporal tensor and the weighted nearest neighbor spatial graph to obtain a traffic accident risk prediction result.

[0005] According to at least one embodiment of the traffic accident risk prediction method of this disclosure, matching the region with regional entities in the traffic accident knowledge graph includes: obtaining the administrative division code or administrative region name to which the region belongs; and based on the administrative division code or administrative region name, retrieving regional entities with the same identifier in the traffic accident knowledge graph to achieve alignment and matching between the region and the regional entities.

[0006] According to at least one embodiment of the traffic accident risk prediction method of this disclosure, a time series tensor containing time and regional dimensions is constructed based on the joint feature vector, including: taking the prediction time as a reference, truncating the joint feature vector in the time dimension according to the sliding window length to obtain an input subsequence containing multiple consecutive times; and structuring the joint feature vector in the input subsequence according to the time order and regional order to generate the time series tensor.

[0007] According to at least one embodiment of the traffic accident risk prediction method of this disclosure, a weighted nearest neighbor spatial graph is constructed based on the similarity between the semantic embedding vectors corresponding to each region in the time series tensor, including: for each region, selecting the K other regions with the highest similarity as the neighbor nodes of the region, where K represents the threshold of the number of neighbors; and determining the weight of the edge between each region and the neighbor node based on the similarity, thereby constructing the weighted nearest neighbor spatial graph.

[0008] According to at least one embodiment of the traffic accident risk prediction method of this disclosure, traffic accident risk prediction is performed based on the temporal tensor and the weighted nearest neighbor spatial graph to obtain a traffic accident risk prediction result, including: performing temporal convolution on the temporal tensor to extract short-term temporal patterns and obtain a first intermediate feature; performing spatial graph convolution on the first intermediate feature based on the weighted nearest neighbor spatial graph to aggregate neighborhood information of semantically similar regions and obtain a second intermediate feature; performing temporal convolution on the second intermediate feature again to obtain a spatiotemporal joint feature; extracting the long-distance temporal dependency of the spatiotemporal joint feature; fusing the long-distance temporal dependency with the last frame joint feature vector in the temporal tensor to obtain a prediction vector; and performing traffic accident risk prediction based on the prediction vector to obtain a traffic accident risk prediction result.

[0009] According to at least one embodiment of the traffic accident risk prediction method of this disclosure, the accident statistical characteristics include: the number of traffic accidents and the number of serious traffic accidents occurring in the corresponding area per unit time.

[0010] According to at least one embodiment of the traffic accident risk prediction method of this disclosure, traffic accident risk prediction is performed based on the prediction vector to obtain traffic accident risk prediction results, including: predicting the prediction vector through two linear prediction heads to obtain the probability of traffic accident occurrence and the severity of traffic accident, respectively; determining that a traffic accident has occurred when the probability of traffic accident occurrence is greater than the probability threshold; and determining whether the traffic accident is a serious traffic accident based on the comparison result between the severity of traffic accident and the severity threshold when a traffic accident has occurred.

[0011] According to another aspect of this disclosure, an electronic device is provided, comprising: a memory storing execution instructions; and a processor executing the execution instructions stored in the memory, causing the processor to perform a traffic accident risk prediction method according to any embodiment of this disclosure.

[0012] According to another aspect of this disclosure, a readable storage medium is provided, wherein executable instructions are stored therein, which, when executed by a processor, are used to implement the traffic accident risk prediction method of any embodiment of this disclosure.

[0013] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements a traffic accident risk prediction method according to any embodiment of this disclosure.

[0014] This disclosure overcomes the problem of missing regional associations caused by data heterogeneity in traditional methods by constructing a traffic accident knowledge graph containing rich semantic information and extracting entity embedding vectors, and combining this with a region matching mechanism to obtain the semantic embedding vectors of regions. By fusing accident statistical features with semantic embedding vectors to construct a temporal tensor, and constructing a weighted nearest neighbor spatial graph based on semantic similarity, it can accurately capture deep semantic dependencies and dynamic spatiotemporal evolution patterns between regions, effectively solving the shortcomings of existing technologies in handling high-dimensional nonlinear interactions and the fusion of external heterogeneous information. This significantly improves the accuracy of traffic accident risk prediction. Attached Figure Description

[0015] The accompanying drawings illustrate exemplary embodiments of the present disclosure and, together with the description thereof, serve to explain the principles of the present disclosure. These drawings are included to provide a further understanding of the present disclosure and are incorporated in and constitute a part of this specification.

[0016] Figure 1 This is a flowchart illustrating a traffic accident risk prediction method according to one embodiment of the present disclosure.

[0017] Figure 2This is a schematic structural block diagram of an electronic device employing a processor-based hardware implementation according to one embodiment of the present disclosure. Detailed Implementation

[0018] The present disclosure will now be described in further detail with reference to the accompanying drawings and examples. It should be understood that the specific examples described herein are for illustrative purposes only and are not intended to limit the scope of the disclosure. Furthermore, it should be noted that, for ease of description, only the parts relevant to the present disclosure are shown in the accompanying drawings.

[0019] It should be noted that, where there is no conflict, the embodiments and features described in this disclosure can be combined with each other. The technical solutions of this disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] Figure 1 A schematic diagram illustrating the overall flow of a traffic accident risk prediction method according to one embodiment of this disclosure is shown. Figure 1 The method shown includes steps S110 to S170. This method can be executed by electronic devices such as mobile phones and tablets.

[0021] S110: Construct a traffic accident knowledge graph based on historical traffic accident data, and determine the embedding vector of each entity in the traffic accident knowledge graph.

[0022] As one possible implementation, TransD or enhanced TransD algorithms can be used to determine the embedding vectors of each entity in the traffic accident knowledge graph, and no restrictions are imposed here.

[0023] S120: Based on historical traffic accident data, determine the statistical characteristics of accidents at different times and in different regions.

[0024] As one possible implementation method, time is distinguished by date. By aggregating and statistically analyzing historical traffic accident data by date and region, the statistical characteristics of accidents in different regions on different dates can be determined.

[0025] As one possible implementation, accident statistics can include the number of accidents and the number of serious accidents. For example, the accident statistics for region n on date t can be represented as a vector or a combination of numbers, specifically including: the number of accidents in region n on date t, and the number of serious accidents in region n on date t.

[0026] S130: Match the region with the regional entities in the traffic accident knowledge graph, and obtain the embedding vector of the matched regional entity as the semantic embedding vector of the region.

[0027] Considering that regions in a knowledge graph exist as entity nodes, these nodes contain rich semantic information and possess learned high-dimensional vector representations. Historical traffic accident data is typically structured tabular data, with regions identified by text fields. When matching regions with regional entities, direct matching often struggles to overcome the naming heterogeneity issues present in multi-source data. This semantic gap prevents the region fields in structured data from accurately mapping to corresponding entity nodes in the knowledge graph. Therefore, one possible implementation involves matching regions with regional entities in the traffic accident knowledge graph, including: obtaining the administrative division code or administrative district name to which the region belongs; and retrieving regional entities with the same identifier from the traffic accident knowledge graph based on the administrative division code or administrative district name to achieve alignment and matching between regions and regional entities.

[0028] S140: The accident statistical features are fused with the semantic embedding vectors of the corresponding regions to obtain the joint feature vectors of each region at different times.

[0029] As one possible implementation, the joint feature vector can be represented as:

[0030] in, This represents the joint eigenvector of region n on date t; This represents the number of accidents in region n on date t. This represents the number of serious accidents in region n on date t. Let T represent the semantic embedding vector of region n; the superscript T indicates transpose. The above formula can be used to concatenate the two scalar features, the number of accidents and the number of serious accidents, with the semantic embedding vector to form a joint feature vector that includes both accident statistical features and semantic features (semantic embedding vector).

[0031] S150: Construct a temporal tensor containing time and regional dimensions based on joint feature vectors.

[0032] By constructing a temporal tensor, discrete region-level features can be integrated into a standard input format suitable for deep learning models (such as spatiotemporal graph neural networks). Specifically, the joint feature vectors generated in step S140 can be structured according to temporal order (e.g., consecutive T days) and spatial range (N regions in total) to form a temporal tensor. Each slice of the temporal tensor corresponds to the "statistical-semantic" joint features of a specific region on a specific date.

[0033] As one possible implementation, a temporal tensor containing time and regional dimensions is constructed based on joint feature vectors. This includes: truncating the joint feature vectors along the time dimension according to the sliding window length, using the prediction time as a reference, to obtain input subsequences containing multiple consecutive time periods. The joint feature vectors in the input subsequences are then arranged in a structured manner according to time and regional order to generate the temporal tensor. The dimensions of the temporal tensor include time, region, and feature dimensions. This implementation, by truncating and structuring joint feature vectors within consecutive time periods using a sliding window mechanism, constructs a standardized input tensor capable of simultaneously representing spatiotemporal dependencies. This provides a unified and complete data foundation for subsequent models to accurately capture short-term temporal fluctuations and spatial correlation patterns of traffic accident risks.

[0034] S160: Construct a weighted nearest neighbor spatial graph based on the similarity between the semantic embedding vectors corresponding to each region in the temporal tensor.

[0035] As one possible implementation method, the formula for calculating similarity is:

[0036] in, Indicates the region With the region The similarity between them; and Representing regions and region semantic embedding vector; The superscript T indicates the L2 norm; the superscript T indicates transpose.

[0037] In a weighted nearest neighbor graph, the weight between two regions is represented by their similarity. One possible implementation involves constructing a weighted nearest neighbor graph based on the similarity between the semantic embedding vectors corresponding to each region in the temporal tensor. This includes: for each region, selecting the K most similar other regions as its neighbor nodes, where K represents a threshold for the number of neighbors; and determining the weights of the edges between each region and its neighbor nodes based on the similarity, thereby constructing the weighted nearest neighbor graph.

[0038] In one example, a weighted nearest neighbor space graph can be represented as a weighted edge list, a symmetric adjacency matrix, or other similar forms. When using a weighted edge list, each edge in the graph is represented as a triple. Where i and j represent the identifiers of two connected region nodes, Let represent the weight between region node i and region node j (i.e., the cosine similarity of their semantic embedding vectors). The weighted edge list contains all directed edges that satisfy the condition "j is a K-nearest neighbor of i". When using a symmetric adjacency matrix, an initial weight matrix W is constructed based on the weighted edge list. If region node j belongs to the neighbor set of region node i, then let Otherwise . Let represent the element in the i-th row and j-th column of the initial weight matrix W. Then, the initial weight matrix W is symmetricized, and further... , to represent the bidirectional semantic adjacency relationship between regions.

[0039] S170: Traffic accident risk prediction is performed based on time series tensors and weighted nearest neighbor spatial graphs, and the traffic accident risk prediction results are obtained.

[0040] As one possible implementation, traffic accident risk prediction can be performed using a traffic accident risk prediction module. This module may include a spatiotemporal convolutional module, a deep model encoder, and a linear prediction head.

[0041] First, the temporal tensor and the weighted nearest neighbor spatial graph are input into the spatiotemporal convolution module. The spatiotemporal convolution module performs joint feature extraction on the input temporal tensor and the weighted nearest neighbor spatial graph to obtain spatiotemporal joint features. For example, the joint feature extraction process includes: performing temporal convolution on the temporal tensor to extract short-term temporal patterns, obtaining a first intermediate feature; performing spatial graph convolution on the first intermediate feature based on the weighted nearest neighbor spatial graph to aggregate neighborhood information of semantically similar regions, obtaining a second intermediate feature; and performing temporal convolution again on the second intermediate feature to obtain the spatiotemporal joint features.

[0042] Subsequently, the spatiotemporal joint features are input into the deep model encoder. The deep model encoder first extracts the long-range temporal dependencies of the spatiotemporal joint features. Then, the long-range temporal dependencies are fused with the joint feature vector of the last frame in the temporal tensor (such as the joint feature vector corresponding to the last date) to obtain the prediction vector. For example, fusing the long-range temporal dependencies with the joint feature vector of the last frame in the temporal tensor to obtain the prediction vector includes: rearranging the spatiotemporal joint features in each region according to temporal order, and then feeding them into a multi-head self-attention Transformer encoder to obtain a high-level temporal representation of each region. The high-level temporal representation is then residually summed with the last frame feature vector of the corresponding region and layer normalized to obtain the prediction vector.

[0043] Finally, the prediction vector is predicted using two linear prediction heads to obtain the probability of a traffic accident and the severity of the traffic accident, respectively. If the probability of a traffic accident is greater than a probability threshold, a traffic accident is determined to have occurred. If a traffic accident is determined to have occurred, based on the comparison between the severity of the traffic accident and a severity threshold, it is determined whether the traffic accident is a serious traffic accident. This implementation only further determines the severity of the traffic accident when the probability of occurrence is positive, which conforms to the semantic prior that severity depends on the occurrence of the accident.

[0044] This disclosure overcomes the problem of missing regional associations caused by data heterogeneity in traditional methods by constructing a traffic accident knowledge graph containing rich semantic information and extracting entity embedding vectors, and combining this with a region matching mechanism to obtain the semantic embedding vectors of regions. By fusing accident statistical features with semantic embedding vectors to construct a temporal tensor, and constructing a weighted nearest neighbor spatial graph based on semantic similarity, it can accurately capture deep semantic dependencies and dynamic spatiotemporal evolution patterns between regions, effectively solving the shortcomings of existing technologies in handling high-dimensional nonlinear interactions and the fusion of external heterogeneous information. This significantly improves the accuracy of traffic accident risk prediction.

[0045] According to further embodiments of this disclosure, an electronic device is also provided. Figure 2 This diagram illustrates a schematic block diagram of an electronic device employing a processor-based hardware implementation according to an embodiment of the present disclosure. The hardware structure of the electronic device of the present disclosure can be implemented using a bus architecture. The bus architecture can include any number of interconnect buses and bridges, depending on the specific application and overall design constraints of the hardware. Bus 1100 connects various circuits including one or more processors 1200, memory 1300, and / or hardware modules. Bus 1100 can also connect various other circuits 1400 such as peripheral devices, voltage regulators, power management circuits, external antennas, etc. Bus 1100 can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Component (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, only one connecting line is used in this figure, but this does not indicate that there is only one bus or one type of bus. The memory 1300 stores a computer program, and when the processor 1200 executes the computer program, the processor 1200 is able to perform the methods described in the above embodiments of this disclosure.

[0046] This disclosure also provides a readable storage medium storing a computer program that, when executed by a processor, is used to implement the methods described above. A "readable storage medium" can be any means capable of containing, storing, communicating, propagating, or transmitting a program for use by or in conjunction with an instruction execution system, apparatus, or device. More specific examples of a readable storage medium include: an electrical connection with one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), fiber optic devices, and portable read-only memory (CDROM), etc.

[0047] This disclosure also provides a computer program product, the methods of which can be implemented wholly or partially through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented wholly or partially as a computer program product. The computer program product includes one or more computer programs or instructions. When the computer program or instructions are loaded and executed, all or part of the processes or functions of this disclosure are performed.

[0048] Computer programs or instructions can be stored in a readable storage medium or transferred from one readable storage medium to another. For example, the computer program or instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The readable storage medium can be any available medium capable of access, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium, such as a floppy disk, hard disk, or magnetic tape; an optical medium, such as a digital video optical disc; or a semiconductor medium, such as a solid-state drive. The computer-readable storage medium can be a volatile or non-volatile storage medium, or it can include both volatile and non-volatile types of storage media.

[0049] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, systems, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0050] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus, and computer program products according to this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0051] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0052] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0053] In the description of this specification, the references to terms such as "one embodiment / mode," "some embodiments / modes," "example," "specific example," or "some examples," etc., refer to specific features, structures, or characteristics described in connection with that embodiment / mode or example, which are included in at least one embodiment / mode or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment / mode or example. Moreover, the specific features, structures, or characteristics described may be combined in any suitable manner in one or more embodiments / modes or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments / modes or examples described in this specification, as well as the features of different embodiments / modes or examples.

[0054] Those skilled in the art should understand that the above embodiments are merely for illustrating the present disclosure and are not intended to limit the scope of the disclosure. Those skilled in the art can make other changes or modifications based on the above disclosure, and these changes or modifications still fall within the scope of the present disclosure.

Claims

1. A method for predicting traffic accident risks, characterized in that, include: A traffic accident knowledge graph is constructed based on historical traffic accident data, and the embedding vector of each entity in the traffic accident knowledge graph is determined. Based on historical traffic accident data, determine the statistical characteristics of accidents in different times and regions; The region is matched with the region entities in the traffic accident knowledge graph, and the embedding vector of the matched region entity is obtained as the semantic embedding vector of the region. The accident statistical features are fused with the semantic embedding vectors of the corresponding regions to obtain the joint feature vectors of each region at different times. Based on the joint feature vector, a temporal tensor containing time and region dimensions is constructed; Based on the similarity between the semantic embedding vectors corresponding to each region in the temporal tensor, a weighted nearest neighbor spatial graph is constructed; and Traffic accident risk prediction is performed based on the time series tensor and the weighted nearest neighbor spatial graph, and the traffic accident risk prediction result is obtained.

2. The traffic accident risk prediction method as described in claim 1, characterized in that, Matching the region with region entities in the traffic accident knowledge graph includes: Obtain the administrative division code or administrative region name to which the region belongs; and Based on the administrative division code or administrative region name, regional entities with the same identifier are retrieved in the traffic accident knowledge graph to achieve alignment and matching between the region and the regional entity.

3. The traffic accident risk prediction method as described in claim 1, characterized in that, Based on the joint feature vector, a temporal tensor containing time and region dimensions is constructed, including: Based on the prediction time, the joint feature vector is truncated along the time dimension according to the sliding window length to obtain an input subsequence containing multiple consecutive times; and The joint feature vectors in the input subsequence are arranged in a structured manner according to time order and region order to generate the temporal tensor.

4. The traffic accident risk prediction method as described in claim 1, characterized in that, Based on the similarity between the semantic embedding vectors corresponding to each region in the temporal tensor, a weighted nearest neighbor spatial graph is constructed, including: For each region, the K other regions with the highest similarity are selected as the neighbor nodes of that region, where K represents the threshold number of neighbors; and The weights of the edges between each region and its neighboring nodes are determined based on the similarity, thereby constructing the weighted nearest neighbor spatial graph.

5. The traffic accident risk prediction method as described in claim 1, characterized in that, Traffic accident risk prediction is performed based on the time-series tensor and the weighted nearest neighbor spatial graph, resulting in traffic accident risk prediction results, including: Temporal convolution is performed on the temporal tensor to extract short-term temporal patterns and obtain a first intermediate feature; based on the weighted nearest neighbor spatial graph, spatial graph convolution is performed on the first intermediate feature to aggregate neighborhood information of semantically similar regions and obtain a second intermediate feature; temporal convolution is performed on the second intermediate feature again to obtain a spatiotemporal joint feature; Extract the long-range temporal dependency of the spatiotemporal joint features; fuse the long-range temporal dependency with the last frame joint feature vector in the temporal tensor to obtain the prediction vector; and Traffic accident risk is predicted based on the predicted vector, and the traffic accident risk prediction result is obtained.

6. The traffic accident risk prediction method as described in claim 5, characterized in that, Based on the prediction vector, traffic accident risk is predicted to obtain the traffic accident risk prediction result, including: The prediction vector is predicted by two linear prediction heads, and the probability of traffic accident and the severity of traffic accident are obtained respectively. If the probability of a traffic accident occurring is greater than a probability threshold, a traffic accident is determined to have occurred; and In cases where a traffic accident has occurred, the severity of the traffic accident is compared with a severity threshold to determine whether the traffic accident is a serious traffic accident.

7. The traffic accident risk prediction method as described in claim 1, characterized in that, The statistical characteristics of the accidents include: the number of traffic accidents and the number of serious traffic accidents occurring in the corresponding area per unit time.

8. An electronic device, characterized in that, include: The memory stores execution instructions; as well as A processor that executes the execution instructions stored in the memory, causing the processor to perform the traffic accident risk prediction method according to any one of claims 1 to 7.

9. A readable storage medium, characterized in that, The readable storage medium stores execution instructions, which, when executed by a processor, are used to implement the traffic accident risk prediction method according to any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the traffic accident risk prediction method according to any one of claims 1 to 7.

Images