Operating vehicle accident high-risk scene portraying method and system

Abstract

The invention relates to a commercial vehicle accident high-risk scene portraying method and system. The method comprises the following steps: acquiring original data; constructing a high-risk scene index system, and defining a risk level; preprocessing the original data based on the high-risk scene index system to obtain preprocessed data; training the self-organizing mapping neural network by utilizing the preprocessed data, and adjusting network node weights by adopting a forgetting suboptimal rule to obtain an initial clustering center; and taking the initial clustering center as a starting point, taking the risk level as an initial K value, and carrying out K-means clustering analysis on the preprocessed data to obtain a commercial vehicle accident high-risk scene portrait result. According to the method, the SOM neural network and the weight updating mechanism with the forgetting factor are introduced to optimize the clustering effect, and the pendulum effect generated in the later period is avoided; through methods of clustering analysis, multiple correspondence analysis and the like, accurate description of an accident high-incidence scene of the commercial vehicle is realized, and a support is provided for comprehensively improving the safety of the road transport vehicle.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of road traffic safety, in particular to a high-risk scene profiling method and system for accidents of commercial vehicles. BACKGROUND

[0002] At present, commercial vehicles are the focus of traffic safety supervision of governments of various countries, and relevant researches are relatively rich, but most of them are based on single subjects such as commercial passenger cars or dangerous chemical transport vehicles, and the research objects show the characteristics of singularity. At the same time, there are few related accident data indicators, and the data dimension is insufficient, lacking multi-dimensional index analysis.

[0003] The existing accident analysis of commercial vehicles focuses on single factors (such as driver behavior or road conditions), lacks systematic research on the multi-dimensional coordination of “people-vehicles-roads-environment”, and the traditional clustering method (such as K-means) is prone to “dimension explosion” in high-dimensional data processing, resulting in low identification accuracy of high-risk scenes and fuzzy scene characteristics. In addition, the existing index system has redundancy or missing, which is difficult to fully reflect the core elements of accident risk.

[0004] In the aspect of safety research of commercial vehicles in the whole process and all elements, governments and regions have taken diversified measures in the safety supervision of commercial vehicles, and domestic and foreign scholars have conducted a large number of researches on the safety and accidents of commercial vehicles, covering many aspects such as running speed characteristics, fault tree analysis, risk assessment model, etc. These researches reveal the risk factors of commercial vehicles in different scenes, such as road environment, driving behavior, etc., but there are few researches on high-risk scenes of commercial vehicle accidents.

[0005] The existing accident profiling researches mostly focus on risk assessment and accident causation in specific vehicle types and scenes, and some researches use scenario construction theory and numerical simulation method to predict the impact and harm of accidents under different conditions. Although the accident profiling research has made certain progress in risk assessment and scenario simulation, it lacks systematic research on high-risk scenes of commercial vehicle accidents. Therefore, it is necessary to characterize the high-risk scenes of commercial vehicle accidents.

[0006] Domestic and foreign scholars have adopted various methods in the aspect of severity of commercial vehicle accidents, covering various Logit models, machine learning models, Bayesian network models and neural network models, etc. These researches reveal the relationship between the severity of accidents and vehicles, drivers, road environment and other factors. Although there are some researches on the severity of commercial vehicle accidents, the overall number is relatively small, and most of them focus on specific scenarios. SUMMARY

[0007] The present application provides a high-risk scene profiling method and system for accidents of commercial vehicles to solve at least one of the above technical problems.

[0008] The technical scheme for solving the above technical problems is as follows: a method for profiling high-risk scenarios of operating vehicle accidents, comprising: S1, obtaining operating vehicle accident original data from multiple channels; and constructing an operating vehicle accident high-risk scenario index system from four dimensions of drivers, vehicles, roads and environments, and defining a risk level for representing the severity of accidents; S2, based on the operating vehicle accident high-risk scenario index system, preprocessing the operating vehicle accident original data to obtain operating vehicle accident preprocessing data; S3, training a self-organizing mapping neural network using the operating vehicle accident preprocessing data, adjusting network node weights using a forgetting suboptimal rule during training, and obtaining an initial clustering center after training is completed; S4, taking the initial clustering center as a starting point and taking the risk level as an initial K value, performing K-means clustering analysis on the operating vehicle accident preprocessing data to obtain an operating vehicle accident high-risk scenario profiling result.

[0009] On the basis of the above technical scheme, the present application can also be improved as follows.

[0010] Further, in the S1, the operating vehicle accident original data is obtained from multiple channels, specifically comprising: Accessing the websites of multiple departments based on a Python crawler script to obtain operating vehicle accident original data.

[0011] Further, in the S1, the operating vehicle accident high-risk scenario index system comprises human factor risk indicators, vehicle factor risk indicators, road factor risk indicators and environmental factor risk indicators; The human factor risk indicators include driver physical condition and driver operation condition, the vehicle factor risk indicators include vehicle condition and vehicle driving state, the road factor risk indicators include road condition, and the environmental factor risk indicators include environmental condition and roadside condition; The driver physical condition includes driver age characteristics and fatigue driving, the driver operation condition includes driver driving condition, driver violation condition and driver handling condition, the vehicle condition includes vehicle model, the vehicle driving state includes vehicle speed and vehicle loading condition, the road condition includes road line type, road lane number, road speed limit and road level, the environmental condition includes visibility and weather condition, and the roadside condition includes roadside natural condition.

[0012] Further, in the S2, the preprocessing comprises one-hot encoding processing, feature conversion processing and principal component analysis processing. Further, in the S2, the preprocessing comprises one-hot encoding processing, feature conversion processing and principal component analysis processing.

[0013] Further, in the S2, further comprising: judging whether the operating vehicle accident preprocessing data meets preset clustering requirements, if not, re-conducting feature conversion processing and the principal component analysis processing on the operating vehicle accident preprocessing data until the operating vehicle accident preprocessing data meets the preset clustering requirements, if yes, executing the S3.

[0014] Further, the S3 specifically comprises: S31, initializing a self-organizing mapping neural network; S32, inputting samples of the operating vehicle accident preprocessing data into the initialized self-organizing mapping neural network, calculating the similarity between the samples and each node in the self-organizing mapping neural network, and selecting the node with the largest similarity as the winning node; S33, determining the winning neighborhood of the winning node according to a winning neighborhood radius function, adjusting the weight of the winning node, and conducting forgetting processing on the weight of the suboptimal node in the winning neighborhood by introducing a weight update mechanism with a forgetting factor; S34, judging whether the self-organizing mapping neural network converges, if not, returning to the S33 for cyclic iteration until the self-organizing mapping neural network converges, if yes, outputting the initial clustering center.

[0015] Further, the winning neighborhood radius function is expressed as: ; Wherein, is the winning neighborhood radius when the iteration number is , is the initial winning neighborhood radius, is a time constant.

[0016] Further, the S4 specifically comprises: S41, taking the risk level as the initial K value; S42, randomly selecting K initial clustering centers of samples in the sample space of the operating vehicle accident preprocessing data; S43, calculating the Euclidean distance between each sample in the sample space of the operating vehicle accident preprocessing data and the selected K initial clustering centers; wherein, the calculation formula of the Euclidean distance is: ; Wherein, is the i-th sample in the sample space of the operating vehicle accident preprocessing data; is the i-th clustering center at the j-th iteration, and is the j-th iteration. ;​​ the Euclidean distance between the i-th sample and the j-th cluster center in the t-th iteration; the Euclidean distance between the i-th sample and the j-th cluster center in the t-th iteration; the feature dimension of the sample; the i-th feature value of the i-th sample; the i-th feature value of the j-th cluster center in the t-th iteration; the i-th feature value of the j-th cluster center in the t-th iteration; S44, according to the Euclidean distance, finding the nearest cluster center of each sample in the sample space of the operating vehicle accident preprocessing data, and assigning the nearest cluster center as the corresponding cluster; wherein the formula for assigning the nearest cluster center as the corresponding cluster is: ; wherein, S45, judging whether the convergence condition is met, if yes, outputting the cluster assignment result, and taking the cluster assignment result as the operating vehicle accident high-risk scene profiling result, if not, updating the initial cluster center of each sample in the sample space of the operating vehicle accident preprocessing data according to the cluster assignment result, and returning to the S42 for loop iteration execution until the convergence condition is met; wherein the formula for updating the initial cluster center of each sample in the sample space of the operating vehicle accident preprocessing data according to the cluster assignment result is: ; wherein, the number of samples in the cluster after the t-th iteration; the updated cluster center in the t-th iteration.

[0017] Further, the S4 further comprises: S46, evaluating the operating vehicle accident high-risk scene profiling result to obtain an evaluation result, if the evaluation result does not meet the preset result, modifying the initial K value, and returning to the S42 for loop iteration execution to obtain the operating vehicle accident high-risk scene profiling result under different K values.​​​​​​​​​​​​​​​​​​ S47. Compare the high-risk scenario profile results of operating vehicle accidents under different K values, and select the best high-risk scenario profile result of operating vehicle accidents as the final high-risk scenario profile result of operating vehicle accidents.

[0018] Based on the above-mentioned method for profiling high-risk scenarios of accidents involving commercial vehicles, the present invention also provides a system for profiling high-risk scenarios of accidents involving commercial vehicles.

[0019] A high-risk scenario profiling system for commercial vehicle accidents includes a processor, a memory, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the aforementioned method for profiling high-risk scenarios for commercial vehicle accidents.

[0020] The beneficial effects of this invention are as follows: This invention provides a method and system for profiling high-risk scenarios of commercial vehicle accidents. It crawls commercial vehicle accident data from multiple channels using data mining technology and establishes an indicator system for high-risk scenarios of commercial vehicle accidents from four dimensions: driver, vehicle, road, and environment. Preprocessing methods are used to remove redundant information and reduce data dimensionality to avoid dimensionality explosion. Cluster analysis is performed based on an improved K-means algorithm. In the improved K-means algorithm, a SOM neural network and a weight update mechanism with a forgetting factor are introduced to optimize the clustering effect and avoid the pendulum effect that may occur later. Through cluster analysis and multiple correspondence analysis, it achieves accurate characterization of high-incidence scenarios of commercial vehicle accidents, reveals the common causes and patterns of accidents, and provides support for identifying significant factors affecting the severity of accidents. This allows for targeted development of accident risk prevention measures and policies related to enterprises and practitioners, preventing similar accidents from recurring, fundamentally mitigating accident risks, and comprehensively improving the safety of road transport vehicles. Attached Figure Description

[0021] Figure 1 This is a flowchart of a high-risk scenario profiling method for commercial vehicle accidents according to the present invention; Figure 2 A schematic diagram of an indicator system for high-risk accident scenarios involving commercial vehicles; Figure 3 This is a structural block diagram of a high-risk scenario profiling system for commercial vehicle accidents according to the present invention. Detailed Implementation

[0022] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0023] like Figure 1 As shown, a method for profiling high-risk accident scenarios of commercial vehicles includes: S1, obtain raw data on commercial vehicle accidents from multiple channels; and construct an indicator system for high-risk scenarios of commercial vehicle accidents from four dimensions: driver, vehicle, road and environment, and define risk levels to characterize the severity of accidents. S2, based on the high-risk scenario index system for operating vehicle accidents, preprocess the original data of the operating vehicle accidents to obtain preprocessed data of operating vehicle accidents. S3, the self-organizing map neural network is trained using the preprocessing data of the operating vehicle accident, and the weights of the network nodes are adjusted by the forgetting suboptimal rule during the training process, and the initial cluster center is obtained after the training is completed. S4. Starting from the initial cluster center and using the risk level as the initial K value, perform K-means clustering analysis on the preprocessed data of the operating vehicle accident to obtain the high-risk scenario profile of the operating vehicle accident.

[0024] This invention presents a method for profiling high-risk scenarios of commercial vehicle accidents. It utilizes data mining technology to crawl accident data from multiple channels and establishes an indicator system for high-risk scenarios based on four dimensions: driver, vehicle, road, and environment. Preprocessing methods are used to remove redundant information and reduce data dimensionality to avoid dimensionality explosion. An improved K-means algorithm is then used for cluster analysis. This improved K-means algorithm incorporates a SOM neural network and a weight update mechanism with a forgetting factor to optimize clustering results and avoid pendulum effects that may occur later. Through cluster analysis and multiple correspondence analysis, the method achieves accurate characterization of high-incidence scenarios of commercial vehicle accidents, revealing common causes and patterns of accidents. This provides support for identifying significant factors affecting the severity of accidents, enabling targeted development of accident risk prevention measures and policies related to enterprises and employees, preventing similar accidents from recurring, mitigating accident risks at their source, and comprehensively improving the safety of road transport vehicles.

[0025] The following is a detailed explanation of each step.

[0026] Preferably, in step S1, obtaining raw accident data of operating vehicles from multiple channels specifically includes: This study used a Python web crawler script to access the websites of multiple departments to obtain raw data on accidents involving commercial vehicles.

[0027] Specifically, data mining techniques were employed, using the Selenium library and Chromedriver in Python to scrape raw data on commercial vehicle accidents. Customized Python web crawlers were written to access government websites, emergency management department platforms, and relevant accident reporting websites, accurately locating and downloading reports containing detailed accident information. Python's data parsing and cleaning capabilities were then used to further extract key fields from the reports, such as accident type, time, location, information on vehicles and drivers involved, descriptions of the accident, and causal analysis.

[0028] Preferred, such as Figure 2 As shown, in S1, the high-risk scenario indicator system for operating vehicle accidents includes human factor risk indicators, vehicle factor risk indicators, road factor risk indicators, and environmental factor risk indicators. The human factor risk indicators include the driver's physical condition and the driver's operation; the vehicle factor risk indicators include the vehicle condition and the vehicle driving status; the road factor risk indicators include the road condition; and the environmental factor risk indicators include the environmental conditions and the roadside conditions. The driver's physical condition includes the driver's age characteristics and fatigue driving; the driver's operation includes the driver's driving situation, driver violations, and driver's handling of situations; the vehicle situation includes the vehicle type; the vehicle driving status includes the vehicle speed and vehicle load; the road situation includes the road alignment, number of lanes, speed limit, and road level; the environmental situation includes visibility and weather conditions; and the roadside situation includes the roadside natural conditions.

[0029] Specifically, following the principles of scientific rigor, systematicity, rationality, independence, and hierarchy, a high-risk scenario indicator system for commercial vehicle accidents is constructed from four dimensions: people, vehicles, roads, and environment. A hierarchical structure design is adopted, subdividing the four dimensions into seven secondary indicators and fifteen tertiary indicators to achieve full coverage of driving operation subjects, vehicle characteristics, infrastructure conditions, and external environmental factors. This maintains the independence between each dimension while establishing logical connections. All terminal indicators are defined using variables that can be directly observed or measured in the commercial vehicle accident database to ensure the feasibility of data analysis.

[0030] In addition, S1 also clarifies different risk levels according to the severity of the accident (general, major, serious, etc.), and this classification will be used for subsequent cluster analysis.

[0031] Preferably, in S2, the preprocessing includes one-hot encoding, feature transformation, and principal component analysis.

[0032] Specifically, one-hot encoding transforms non-numerical, discrete categorical information such as weather ("sunny / rainy / snowy") and road alignment ("straight / curved / slope") into a numerical matrix encoding method: a 1 is recorded if the feature exists, and a 0 is recorded if it doesn't. Feature transformation standardizes numerical indicators to make data with different dimensions comparable. For example, data with large differences in numerical ranges, such as driver age and vehicle speed, are standardized and scaled to a range of 0-1 or a mean of 0. Principal component analysis (PCA) aims to reduce dimensionality (when there are too many features, it evaluates the variance contribution of all variables in the original dataset, uses statistical screening to remove indicators with weak explanatory power, reduces dimensionality, retains core information, and avoids the curse of dimensionality). PCA is used to remove redundant information and indicators with weak explanatory power from the original data of commercial vehicle accidents, retaining principal components with high cumulative variance contribution rates. This reduces the dimensionality of data with poor explanatory power in the previous human-vehicle-road-environment profiling system, avoids dimensionality explosion, and simplifies subsequent clustering calculations.

[0033] Preferably, step S2 further includes: determining whether the preprocessed data of the operating vehicle accident meets the preset clustering requirements; if not, then re-performing the feature transformation processing and principal component analysis processing of the preprocessed data of the operating vehicle accident until the preprocessed data of the operating vehicle accident meets the preset clustering requirements; if so, then executing step S3.

[0034] Specifically, check whether the preprocessed data of operating vehicle accidents meets the requirements of subsequent clustering (such as silhouette coefficient, CH index, etc.). If it does not meet the requirements, the feature transformation and PCA dimensionality reduction are readjusted, such as changing the standardization method and adjusting the dimensionality reduction dimension, so as to finally obtain operating vehicle accident preprocessed data suitable for subsequent clustering analysis.

[0035] Preferably, S3 specifically includes: S31, Initialize the self-organizing map neural network; S32, input the sample of the preprocessing data of the operating vehicle accident into the initialized self-organizing map neural network, calculate the similarity between the sample and each node in the self-organizing map neural network, and select the node with the highest similarity as the winning node. S33, determine the winning neighborhood of the winning node according to the winning neighborhood radius function; adjust the weight of the winning node, and perform forgetting processing on the weight of the second-best node in the winning neighborhood by introducing a weight update mechanism with a forgetting factor. S34, determine whether the self-organizing map neural network has converged. If not, return to S33 and iterate until the self-organizing map neural network converges. If it has converged, output the initial cluster center.

[0036] Specifically, a self-organizing map (SOM) neural network is trained using preprocessing data of commercial vehicle accidents to automatically select the number of clusters and initial cluster centers, perform preliminary dimensionality reduction and grouping of high-dimensional data, avoid dimensionality explosion, achieve high-dimensional space compression, and generate an initial partition that preserves topology.

[0037] The main steps in training the SOM neural network are: initializing the node network and sequentially inputting the preprocessed data of commercial vehicle accidents; calculating the similarity (Euclidean distance) between the node and all other nodes; finding the optimal node; and adjusting the weights of the optimal node and its neighboring nodes based on the input data. This process is repeated until the node features essentially no longer change (i.e., the model converges), at which point the initial cluster centers are obtained. During this iterative process, the radius of the winning neighborhood and the learning rate (the degree of weight adjustment) decrease continuously with each iteration, achieving a coarse-to-fine effect.

[0038] The radius function of the winning neighborhood is the number of iterations. The decreasing function, according to the weight adjustment rules, allows only nodes within the neighborhood of the winning node to have their weights updated. Its expression is: ; in, For the number of iterations The winning neighborhood radius at that time This is the initial winning neighborhood radius (usually set to a large value). It is a time constant that controls the rate at which the radius decays.

[0039] The function defined by the learning rate should change with the number of iterations. As the value increases, it decreases, affecting the convergence speed of the model. Its expression is: ; In the formula, For the number of iterations Learning rate at time The initial learning rate, This is the time constant for the learning rate decay.

[0040] To avoid the pendulum effect that may occur later, this invention utilizes a forgetting suboptimal rule. By introducing a weight update mechanism with a forgetting factor, during the training of the SOM neural network model, each time samples of preprocessed data from commercial vehicle accidents are input and their similarity is compared, the weight vectors are adjusted. The weights of the suboptimal nodes are then subject to forgetting. During the update process, the adaptive adjustment of the suboptimal node's weights to the current input sample is deliberately reduced, and they may even be reset or weakened, thereby widening the difference between the top two neurons. This not only enhances the competitive advantage of the winning neuron in attracting similar samples but also increases the diversity of the suboptimal node's weight vector. This improvement enhances the applicability of the SOM neural network in profiling high-risk scenarios of commercial vehicles and mitigates the pendulum effect problem that easily occurs in the later stages of SOM neural network iteration.

[0041] After the SOM neural network model is trained and the initial cluster centers are determined, the preprocessed data of operating vehicle accidents are further refined into groups. The K-means clustering algorithm (also known as K-means clustering algorithm) is used to ensure that the scene similarity within the group is the highest and the difference between the groups is the greatest, so as to obtain the high-risk scene profile of operating vehicle accidents.

[0042] Preferably, S4 specifically includes S41 to S45: S41, using the aforementioned risk level as the initial K value.

[0043] Specifically, the initial K value is determined based on the risk level defined in S1 (if the risk level includes general, significant, and major, then K=3), which serves as the initial number of clusters for cluster analysis.

[0044] S42, randomly select the initial cluster centers of K samples in the sample space of the preprocessing data of the operating vehicle accident.

[0045] Specifically, assuming the sample set of pre-processing data for commercial vehicle accidents is of the following size: The sample set is Randomly select K initial cluster centers from them. ,in .

[0046] S43, calculate the Euclidean distance between each sample in the sample space of the preprocessing data for the operating vehicle accident and the selected K initial cluster centers; wherein, the formula for calculating the Euclidean distance is: ; In the formula, The first in the sample space of the preprocessing data for the operating vehicle accidents One sample; For the first During the nth iteration There are cluster centers, and ; For the first The sample and the first During the nth iteration The Euclidean distance between the cluster centers; The feature dimensions of the sample (each feature dimension of each sample corresponds to an initial cluster center). For the first The first sample One eigenvalue; For the first During the nth iteration The first cluster center The eigenvalue (the th feature value among all samples in this cluster) (The average value of each feature).

[0047] Specifically, S43 assigns data points to the nearest cluster. It establishes a metric relationship between the observation vector and the centroid node and uses the minimum neighborhood criterion to iteratively assign data points to the optimal cluster.

[0048] S44, based on the Euclidean distance, find the nearest cluster center of each sample in the sample space of the preprocessing data of the operating vehicle accident, and assign the nearest cluster center to the corresponding cluster; wherein, the formula for assigning the nearest cluster center to the corresponding cluster is: ; In the formula, For the first During the nth iteration A sample set with cluster centers; For the first During the next iteration All other cluster centers except those in the cluster; S45, determine whether the convergence condition is met. If yes, output the clustering assignment result and use it as the high-risk scenario profile result for operating vehicle accidents. If not, update the initial cluster centers of each sample in the sample space of the preprocessed data of operating vehicle accidents according to the clustering assignment result, and return to S42 for iterative execution until the convergence condition is met. The formula for updating the initial cluster centers of each sample in the sample space of the preprocessed data of operating vehicle accidents according to the clustering assignment result is: ; In the formula, For the first During the nth iteration The first in the cluster One sample; for The number of samples in the cluster after the next iteration; For the first The cluster centers are updated in the next iteration.

[0049] The convergence condition is any one of the following three conditions: (1) Cluster centers are stable: all cluster centers no longer change, i.e. For all (2) The objective function changes little: the change in the sum of squares (Inertia) within the cluster is less than the threshold. (3) Reaching the maximum number of iterations: Set the maximum number of iterations. T max (like T max = 100).

[0050] Furthermore, the objective function is expressed as: ; In the formula, This represents the objective function value.

[0051] Preferably, S4 further includes: S46, evaluate the high-risk scenario profile of the operating vehicle accident and obtain the evaluation result. If the evaluation result does not meet the preset result, modify the initial K value and return to S42 for iterative execution to obtain the high-risk scenario profile of the operating vehicle accident under different K values. S47. Compare the high-risk scenario profile results of operating vehicle accidents under different K values, and select the best high-risk scenario profile result of operating vehicle accidents as the final high-risk scenario profile result of operating vehicle accidents.

[0052] Specifically, the clustering assignment results include cluster partitioning. Cluster Center and objective function value The K value is evaluated using metrics such as silhouette coefficient, CH index, and DBI. If the final result is unsatisfactory, the K value is modified to improve the cluster analysis. The evaluation metrics of the clustering results produced by different K values ​​are compared to obtain the final clustering result.

[0053] Table 1 below shows the clustering analysis index data of SOM+k-means clustering in this invention, compared with existing k-means clustering and k-means++ clustering.

[0054] Table 1: Cluster Analysis Indicators

[0055] Among them, the silhouette coefficient measures the similarity between a sample and its cluster. The closer the value is to 1, the more reasonable the clustering is; the CH index: the larger the value, the higher the similarity within the cluster and the greater the difference between the clusters; the DBI: the smaller the value, the better the clustering effect.

[0056] In this invention, the overall judgment criterion between the evaluation results and the preset results is as follows: if the indicators meet the standards (e.g., silhouette coefficient > 0.1, CH index > 20), the clustering effect is good; otherwise, return to the "parameter adjustment" step and re-optimize the learning rate of SOM, neighborhood radius, or K value of K-means.

[0057] S4 of this invention is an iterative process of K-means clustering. Preprocessed data of operational vehicle accidents is used as input samples. The distance between each sample and each cluster center is calculated, and the sample is assigned to the nearest cluster. The mean of each cluster is recalculated, and the cluster centers are updated. This process is repeated until the cluster centers stabilize or the maximum number of iterations is reached, thus obtaining the clustering assignment result. If the clustering assignment result is not ideal, the K value is readjusted. The optimal K value is determined in conjunction with the business scenario, and the features of each cluster are extracted. Dimensions with high contribution rates within the clusters are analyzed to form a high-risk scenario profile.

[0058] In this embodiment, the typical high-risk scenarios include: (1) general accident scenario: driver age ≥ 40 years old + curve + bad weather + high-risk roadside; (2) major accident scenario: fatigued driving + overload / overcrowding + high-risk roadside; (3) serious accident scenario: hazardous chemical vehicle + high-speed curve + rainy / foggy weather.

[0059] Profiles of high-risk accident scenarios for commercial vehicles can only reveal the overall characteristics. Due to the differences in vehicle characteristics and operating characteristics between commercial passenger vehicles and commercial freight vehicles, the specific scenarios in which accidents occur will also differ. Therefore, after classifying the two types of vehicles, further analysis was conducted, and the typical high-risk scenarios obtained include: (1) General accident scenarios for commercial passenger vehicles: long-term straight driving + distracted driving; (2) Major accident scenarios for commercial passenger vehicles: straight highway + overloading + violation of traffic regulations + severe weather; (3) Serious accident scenarios for commercial passenger vehicles: low visibility at night + multi-lane highway + fatigued driving; (4) General accident scenarios for commercial freight vehicles: long-term straight driving + distracted driving; (5) Serious accident scenarios for commercial freight vehicles: fatigued driving + overloading + intersections of low-grade roads or complex road sections; (6) Serious accident scenarios for commercial freight vehicles: low visibility rain, snow and fog + high-speed curves + speeding.

[0060] The exemplary results of profiling high-risk accident scenarios of commercial vehicles using the method of the present invention are shown in Tables 2 and 3 below.

[0061] Table 2: Results of High-Risk Scenarios for Passenger Buses

[0062] Table 3: Results of High-Risk Scene Profiling for Trucks

[0063] Based on the above-mentioned method for profiling high-risk scenarios of accidents involving commercial vehicles, the present invention also provides a system for profiling high-risk scenarios of accidents involving commercial vehicles.

[0064] like Figure 3 As shown, a high-risk scenario profiling system for commercial vehicle accidents includes a processor, a memory, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the aforementioned method for profiling high-risk scenarios for commercial vehicle accidents.

[0065] In other words, the high-risk scenario profiling system for commercial vehicle accidents in this embodiment of the invention may include, but is not limited to: a processor and a memory; the memory is used to store computer programs; the processor is used to execute the high-risk scenario profiling method for commercial vehicle accidents of the present invention by calling the computer programs.

[0066] In one optional embodiment, a high-risk scenario profiling system for commercial vehicle accidents is provided, such as... Figure 3 As shown. Figure 3 The illustrated high-risk scenario profiling system for commercial vehicle accidents includes a processor and a memory. The processor and memory are connected, for example, via a bus. Optionally, the high-risk scenario profiling system for commercial vehicle accidents may also include a transceiver, which can be used for data interaction between the system and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver is not limited to one unit, and the structure of this high-risk scenario profiling system for commercial vehicle accidents does not constitute a limitation on the embodiments of the present invention.

[0067] The processor can be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), a PLC (Programmable Controllers), a FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this invention. The processor can also be a combination that implements computational functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0068] A bus can include a pathway for transmitting information between the aforementioned components. The bus can be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 3 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0069] The memory may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited to these.

[0070] The memory stores application code (computer program) that executes the present invention, and its execution is controlled by a processor. The processor executes the application code stored in the memory to implement the content shown in the foregoing method embodiments.

[0071] Among them, the high-risk scenario profiling system for commercial vehicle accidents can also be a terminal device. The terminal device can be any device that can install applications, including at least one of the following: smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, smart TV, and smart in-vehicle device.

[0072] It should be noted that, Figure 3 The illustrated high-risk scenario profiling system for commercial vehicle accidents is merely an example and should not be construed as limiting the functionality or scope of the embodiments of the present invention. In one feasible embodiment, the processor's hardware environment is as follows: CPU: 12th Gen Intel(R) Core(TM) i5-12400 2.50 GHz; RAM: 32GB; Windows 10, 64-bit operating system. Software environment: Data preprocessing used WPS Excel and SPSS; the algorithm is implemented using Python 3, and the programming software is PyCharm 2024.3.1.1 x64.

[0073] This invention achieves accident scene profiling through improved cluster analysis of accident-related data: Based on a hierarchical high-risk scene profiling index system for commercial vehicles, it uses SOM neural network + k-means clustering algorithm to profile commercial vehicle accident scenes, conducts in-depth research on high-risk accident scenes of commercial vehicles, expands the explanatory variable system for accidents, and improves the cluster analysis algorithm to make the causal analysis of high-risk accident scenes of commercial vehicles more scientific and systematic, which can provide support for the formulation of relevant laws and regulations for commercial vehicles.

[0074] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for profiling high-risk accident scenarios involving commercial vehicles, characterized in that, include: S1, obtain raw data on accidents involving commercial vehicles from multiple channels; In addition, an indicator system for high-risk scenarios of commercial vehicle accidents is constructed from four dimensions: drivers, vehicles, roads and environment, and risk levels are defined to characterize the severity of accidents; S2, based on the high-risk scenario index system for operating vehicle accidents, preprocess the original data of the operating vehicle accidents to obtain preprocessed data of operating vehicle accidents. S3, the self-organizing map neural network is trained using the preprocessing data of the operating vehicle accident, and the weights of the network nodes are adjusted by the forgetting suboptimal rule during the training process, and the initial cluster center is obtained after the training is completed. S4. Starting from the initial cluster center and using the risk level as the initial K value, perform K-means clustering analysis on the preprocessed data of the operating vehicle accident to obtain the high-risk scenario profile of the operating vehicle accident.

2. The method for profiling high-risk accident scenarios of commercial vehicles according to claim 1, characterized in that, In S1, raw accident data of commercial vehicles is obtained from multiple channels, specifically including: This study used a Python web crawler script to access the websites of multiple departments to obtain raw data on accidents involving commercial vehicles.

3. The method for profiling high-risk accident scenarios of commercial vehicles according to claim 1, characterized in that, In S1, the high-risk scenario indicator system for operating vehicle accidents includes human factor risk indicators, vehicle factor risk indicators, road factor risk indicators, and environmental factor risk indicators. The human factor risk indicators include the driver's physical condition and the driver's operation; the vehicle factor risk indicators include the vehicle condition and the vehicle driving status; the road factor risk indicators include the road condition; and the environmental factor risk indicators include the environmental conditions and the roadside conditions. The driver's physical condition includes the driver's age characteristics and fatigue driving; the driver's operation includes the driver's driving situation, driver violations, and driver's handling of situations; the vehicle situation includes the vehicle type; the vehicle driving status includes the vehicle speed and vehicle load; the road situation includes the road alignment, number of lanes, speed limit, and road level; the environmental situation includes visibility and weather conditions; and the roadside situation includes the roadside natural conditions.

4. The method for profiling high-risk accident scenarios of commercial vehicles according to claim 1, characterized in that, In S2, the preprocessing includes one-hot encoding, feature transformation, and principal component analysis.

5. The method for profiling high-risk accident scenarios of commercial vehicles according to claim 4, characterized in that, S2 further includes: determining whether the preprocessing data of the operating vehicle accident meets the preset clustering requirements; if not, re-processing the preprocessing data of the operating vehicle accident with feature transformation and principal component analysis until the preprocessing data of the operating vehicle accident meets the preset clustering requirements; if so, then executing S3.

6. The method for profiling high-risk accident scenarios of commercial vehicles according to claim 1, characterized in that, S3 specifically includes: S31, Initialize the self-organizing map neural network; S32, input the sample of the preprocessing data of the operating vehicle accident into the initialized self-organizing map neural network, calculate the similarity between the sample and each node in the self-organizing map neural network, and select the node with the highest similarity as the winning node. S33, determine the winning neighborhood of the winning node according to the winning neighborhood radius function; adjust the weight of the winning node, and perform forgetting processing on the weight of the second-best node in the winning neighborhood by introducing a weight update mechanism with a forgetting factor. S34, determine whether the self-organizing map neural network has converged. If not, return to S33 and iterate until the self-organizing map neural network converges. If it has converged, output the initial cluster center.

7. The method for profiling high-risk accident scenarios of commercial vehicles according to claim 1, characterized in that, The winning neighborhood radius function is expressed as: ； in, For the number of iterations The winning neighborhood radius at that time The initial winning neighborhood radius, is the time constant.

8. The method for profiling high-risk accident scenarios of commercial vehicles according to claim 1, characterized in that, S4 specifically includes: S41, using the aforementioned risk level as the initial K value; S42, randomly select the initial cluster centers of K samples in the sample space of the preprocessing data of the operating vehicle accident; S43, calculate the Euclidean distance between each sample in the sample space of the preprocessing data for the operating vehicle accident and the selected K initial cluster centers; wherein, the formula for calculating the Euclidean distance is: ； In the formula, The first in the sample space of the preprocessing data for the operating vehicle accidents One sample; For the first During the nth iteration There are cluster centers, and ; For the first The sample and the first During the nth iteration The Euclidean distance between the cluster centers; The feature dimension of the sample; For the first The first sample One eigenvalue; For the first During the nth iteration The first cluster center One eigenvalue; S44, based on the Euclidean distance, find the nearest cluster center of each sample in the sample space of the preprocessing data of the operating vehicle accident, and assign the nearest cluster center to the corresponding cluster; wherein, the formula for assigning the nearest cluster center to the corresponding cluster is: ； In the formula, For the first During the nth iteration A sample set with cluster centers; For the first During the next iteration All other cluster centers except those in the cluster; S45, determine whether the convergence condition is met. If yes, output the clustering assignment result and use it as the high-risk scenario profile result for operating vehicle accidents. If not, update the initial cluster centers of each sample in the sample space of the preprocessed data of operating vehicle accidents according to the clustering assignment result, and return to S42 for iterative execution until the convergence condition is met. The formula for updating the initial cluster centers of each sample in the sample space of the preprocessed data of operating vehicle accidents according to the clustering assignment result is: ； In the formula, For the first During the nth iteration The first in the cluster One sample; for The number of samples in the cluster after the next iteration; For the first The cluster centers are updated in the next iteration.

9. The method for profiling high-risk accident scenarios of commercial vehicles according to claim 8, characterized in that, S4 further includes: S46, evaluate the high-risk scenario profile of the operating vehicle accident and obtain the evaluation result. If the evaluation result does not meet the preset result, modify the initial K value and return to S42 for iterative execution to obtain the high-risk scenario profile of the operating vehicle accident under different K values. S47. Compare the high-risk scenario profile results of operating vehicle accidents under different K values, and select the best high-risk scenario profile result of operating vehicle accidents as the final high-risk scenario profile result of operating vehicle accidents.

10. A profiling system for high-risk accident scenarios involving commercial vehicles, characterized in that, The system includes a processor, a memory, and a computer program stored in the memory, wherein the computer program, when executed by the processor, implements the high-risk scenario profiling method for operating vehicle accidents as described in any one of claims 1 to 9.

Images