Vehicle safety assessment method and device based on accident data

Abstract

The invention provides a vehicle safety assessment method and device based on accident data, and the method comprises the steps: obtaining a historical accident data set corresponding to a historical vehicle accident of a target vehicle, carrying out the data fusion of the historical accident data set, and generating target accident data corresponding to the historical vehicle accident; performing clustering analysis on the plurality of target accident data, and identifying at least one risk accident scene; and for each risk accident scene, simulating the risk accident scene by using the simulation model corresponding to the target vehicle so as to evaluate the performance of the safety configuration of the target vehicle in the risk accident scene, and determining a safety configuration evaluation result of the target vehicle. Through the method and the device, the safety performance of the vehicle safety configuration under the extreme working condition can be evaluated before a real accident occurs in a large scale, and the pertinence, the foresight and the real coverage degree of safety evaluation are remarkably improved.

Description

Technical Field

[0001] This application relates to the field of vehicle safety assessment technology, and in particular to a vehicle safety assessment method and apparatus based on accident data. Background Technology

[0002] With the rapid development of the automotive industry, vehicle safety performance has become a core focus in the research and development process. Traditional safety development mainly relies on standardized laboratory tests, such as new vehicle evaluation procedures (NCAPs) and other standardized crash tests. These tests evaluate safety features such as vehicle stability systems and automatic emergency braking systems by simulating crash conditions in fixed scenarios, such as frontal and side collisions. In addition, limited real-world track testing is usually conducted to supplement the shortcomings of laboratory testing.

[0003] However, real-world traffic accident scenarios are highly complex and unpredictable, involving varying road conditions, weather factors, vehicle interaction patterns, and differences in driver behavior. While standard laboratory tests can cover some typical conditions, they cannot fully simulate high-risk scenarios such as multi-vehicle collisions in extreme weather, accidents in complex road topologies, or sudden interactions between road users. Due to the lack of systematic collection and analysis of real-world accident data, traditional safety development methods struggle to dynamically identify these potentially high-risk conditions, causing vehicle safety assessments to often lag behind changes in actual road risks. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide a vehicle safety assessment method and device based on accident data. By deeply mining massive multi-dimensional accident data, it actively identifies potential high-risk scenarios that are not covered by standard tests, and conducts simulation tests on the identified high-risk scenarios in the vehicle digital model to quantify the effectiveness boundary of safety configuration. It can assess the safety performance of vehicle safety configuration under extreme conditions before large-scale real accidents occur, significantly improving the pertinence, foresight and real-world coverage of safety assessment.

[0005] In a first aspect, embodiments of this application provide a vehicle safety assessment method based on accident data, the vehicle safety assessment method comprising: Obtain the historical accident dataset corresponding to the historical accidents of the target vehicle, and perform data fusion on the historical accident dataset to generate the target accident data corresponding to the historical accidents of the vehicle. Cluster analysis was performed on multiple target accident data to identify at least one risk accident scenario; For each risk incident scenario, the simulation model corresponding to the target vehicle is used to simulate the risk incident scenario in order to evaluate the performance of the safety configuration of the target vehicle under the risk incident scenario and determine the safety configuration evaluation result of the target vehicle.

[0006] Furthermore, after determining the safety configuration assessment results of the target vehicle, the vehicle safety assessment method further includes: The safety configuration evaluation results are input into a pre-built vehicle engineering knowledge graph corresponding to the target vehicle to determine the safety optimization suggestions corresponding to the safety configuration.

[0007] Furthermore, the historical accident dataset includes accident scene investigation text, accident scene images, vehicle operation sequence data, and weather data related to the accident. Data fusion is performed on the historical accident dataset to generate target accident data corresponding to the vehicle's historical accidents, including: Using a large language model, named entity recognition and relation extraction are performed on the accident scene investigation text to determine the accident semantic feature set corresponding to the vehicle's historical accidents; The image detection model is used to identify vehicle damage in the accident scene images to determine the vehicle damage feature set corresponding to the vehicle's historical accidents. The vehicle runtime sequence data and the accident weather data are aligned with a time window and associated with the accident semantic feature set and the vehicle damage feature set to obtain the target accident data.

[0008] Furthermore, the cluster analysis of multiple target accident data to identify at least one risk accident scenario includes: Each target accident data is converted into a corresponding high-dimensional feature vector, and a clustering algorithm is used to cluster each high-dimensional feature vector to obtain accident feature clusters with similar accident feature combinations. For each accident feature cluster, the frequency of occurrence of the accident feature cluster in multiple historical accident datasets is counted, and the risk score of the accident feature cluster is determined by combining the accident impact results of the historical accidents of the corresponding vehicles. When the risk score is greater than or equal to a preset score threshold, the cluster of accident features is taken as the risk accident scenario.

[0009] Furthermore, the step of simulating the risky accident scenario using the simulation model corresponding to the target vehicle to evaluate the performance of the target vehicle's safety configuration under the risky accident scenario and determine the safety configuration evaluation result of the target vehicle includes: The risk incident scenario is converted into the initial conditions corresponding to the simulation environment, and the simulation model is called to run the simulation in the simulation environment to determine the intervention result of the safety configuration and the incident simulation result; Using the Monte Carlo simulation algorithm, at least one variable in the initial conditions is randomly adjusted within a preset range, and multiple repeated simulations are performed. The effective intervention rate and failure probability of the safety configuration under the risk incident scenario are statistically analyzed to obtain the safety configuration evaluation result corresponding to the safety configuration.

[0010] Furthermore, the statistical analysis of the effective intervention rate and failure probability of the security configuration in this risk incident scenario includes: The results of multiple accident simulations were analyzed to determine the effective number of times the safety configuration was activated and achieved control within a preset time window, as well as the number of times the safety configuration failed to activate beyond the preset time window and failed to activate beyond the preset time window. The effective intervention rate is determined based on the number of effective interventions and the number of simulations, and the failure probability is determined based on the number of failures and the number of simulations.

[0011] Secondly, embodiments of this application also provide a vehicle safety assessment device based on accident data, the vehicle safety assessment device comprising: The accident data acquisition module is used to acquire the historical accident dataset corresponding to the historical accidents of the target vehicle, and to perform data fusion on the historical accident dataset to generate the target accident data corresponding to the historical accidents of the vehicle. The risk incident scenario identification module is used to perform cluster analysis on multiple target incident data to identify at least one risk incident scenario. The safety assessment module is used to simulate each risk accident scenario using the simulation model corresponding to the target vehicle, in order to evaluate the performance of the safety configuration of the target vehicle under the risk accident scenario and determine the safety configuration assessment result of the target vehicle.

[0012] Furthermore, the vehicle safety assessment device also includes an optimization suggestion determination module. After determining the safety configuration assessment result of the target vehicle, the optimization suggestion determination module is used to: The safety configuration evaluation results are input into a pre-built vehicle engineering knowledge graph corresponding to the target vehicle to determine the safety optimization suggestions corresponding to the safety configuration.

[0013] Thirdly, embodiments of this application also provide an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, the steps of the vehicle safety assessment method based on accident data described above are performed.

[0014] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps of the vehicle safety assessment method based on accident data as described above.

[0015] This application provides a vehicle safety assessment method and apparatus based on accident data. First, a historical accident dataset corresponding to the historical accidents of the target vehicle is obtained, and the historical accident dataset is fused to generate target accident data corresponding to the historical accidents of the vehicle. Then, cluster analysis is performed on multiple target accident datasets to identify at least one risky accident scenario. Finally, for each risky accident scenario, a simulation model corresponding to the target vehicle is used to simulate the risky accident scenario to evaluate the performance of the safety configuration of the target vehicle under the risky accident scenario and determine the safety configuration assessment result of the target vehicle.

[0016] This application generates high-fidelity target accident data by acquiring historical accident datasets of target vehicles and fusing multi-source data. It then performs cluster analysis on multiple target accident datasets to automatically identify high-frequency or high-hazard risk accident scenarios. Finally, it uses simulation models to evaluate the performance of vehicle safety configurations under these risk scenarios, ultimately outputting quantitative evaluation results. Compared to traditional safety assessment methods that rely on fixed standard tests, this application proactively identifies potential high-risk scenarios not covered by standard tests through in-depth mining of massive, multi-dimensional accident data. These identified high-risk scenarios are then simulated and tested in vehicle digital models, quantifying the effectiveness boundaries of safety configurations. This allows for the evaluation of vehicle safety configurations under extreme conditions before large-scale real-world accidents occur, significantly improving the relevance, foresight, and real-world coverage of safety assessments.

[0017] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart illustrating a vehicle safety assessment method based on accident data provided in this application embodiment; Figure 2 This is one of the structural schematic diagrams of a vehicle safety assessment device based on accident data provided in an embodiment of this application; Figure 3 This is a second schematic diagram of the structure of a vehicle safety assessment device based on accident data provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. Based on the embodiments of this application, every other embodiment obtained by those skilled in the art without inventive effort falls within the scope of protection of this application.

[0021] First, the applicable scenarios for this application will be introduced. This application can be applied to the field of vehicle safety assessment technology.

[0022] With the rapid development of the automotive industry, vehicle safety performance has become a core focus in the research and development process. Research has found that traditional safety development primarily relies on standardized laboratory tests, such as new vehicle evaluation procedures (NCAPs) and other standardized crash tests. These tests assess safety features like vehicle stability systems and automatic emergency braking systems by simulating fixed collision scenarios, such as frontal and side impacts. In addition, limited real-world testing is typically conducted to supplement laboratory testing.

[0023] However, real-world traffic accident scenarios are highly complex and unpredictable, involving varying road conditions, weather factors, vehicle interaction patterns, and differences in driver behavior. While standard laboratory tests can cover some typical conditions, they cannot fully simulate high-risk scenarios such as multi-vehicle collisions in extreme weather, accidents in complex road topologies, or sudden interactions between road users. Due to the lack of systematic collection and analysis of real-world accident data, traditional safety development methods struggle to dynamically identify these potentially high-risk conditions, causing vehicle safety assessments to often lag behind changes in actual road risks.

[0024] Based on this, the embodiments of this application provide a vehicle safety assessment method based on accident data, which can assess the safety performance of vehicle safety configurations under extreme conditions before large-scale real accidents occur, significantly improving the pertinence, foresight and real-world coverage of the safety assessment.

[0025] Please see Figure 1 , Figure 1 This is a flowchart illustrating a vehicle safety assessment method based on accident data, provided as an embodiment of this application. Figure 1 As shown in the embodiments of this application, the vehicle safety assessment method includes: S101, obtain the historical accident dataset corresponding to the historical accidents of the target vehicle, and perform data fusion on the historical accident dataset to generate the target accident data corresponding to the historical accidents of the vehicle.

[0026] Here, the target vehicle refers to the specific vehicle model that needs to be assessed for safety. Vehicle history accidents refer to the traffic accident records that have occurred during the actual use of the target vehicle, which may include various accident events such as collisions, rollovers, and rear-end collisions; this application does not specifically limit this. Historical accident dataset refers to a collection of multi-source heterogeneous data related to the target vehicle's historical accidents, which typically includes text, images, sensor time-series data, and environmental information; this application does not specifically limit this. Target accident data refers to standard accident data units generated from the historical accident dataset after data fusion, used for subsequent analysis, and includes comprehensive features such as semantics, visual characteristics, operating status, and environment.

[0027] Regarding step S101 above, in specific implementation, historical accident data related to the historical accidents that occurred with the target vehicle is obtained, and then the historical accident data is fused, aligned, correlated and integrated to form a unified structured representation of the target accident data, so as to improve data integrity and usability.

[0028] According to the embodiments provided in this application, the historical accident dataset centrally includes accident scene investigation text, accident scene images, vehicle operation sequence data, and accident weather data. Specifically, the accident scene investigation text refers to a textual description of the accident, which may include information such as time, geographical location, road type (e.g., highway, urban road), traffic conditions, as well as accident reports and on-site records from insurance surveyors, such as natural language descriptions like "the front left of the vehicle collided with the guardrail." Accident scene images refer to photographs or surveillance video frames taken at the accident scene, used to visually display vehicle damage, road markings, obstacle locations, etc. Vehicle operation sequence data refers to time-series signals collected through the vehicle's ECU or OBD interface, extracted from the accident vehicle's event data recorder, including vehicle speed, acceleration, engine speed, steering wheel angle, brake pedal opening, accelerator opening, etc. Accident weather data refers to environmental parameters such as local temperature, precipitation, wind speed, visibility, and road surface temperature at the time of the vehicle's historical accident.

[0029] Specifically, regarding step S101 above, data fusion is performed on the historical accident dataset to generate target accident data corresponding to the vehicle's historical accidents, including: Step 1011: Use a large language model to perform named entity recognition and relation extraction on the accident scene investigation text to determine the accident semantic feature set corresponding to the vehicle's historical accidents.

[0030] Regarding step 1011 above, in specific implementation, for the accident scene investigation text, the accident scene investigation text is input into a pre-trained large language model. The large language model is used to perform named entity recognition and relation extraction on the accident scene investigation text, identifying entity labels and entity relationships such as "collision object" (guardrail, vehicle, etc.), "damaged parts" (left front longitudinal beam, A-pillar, etc.), and "mechanical condition" (airbag deployment, brake failure, etc.), and converting them into structured feature labels to obtain the accident semantic feature set corresponding to the vehicle's historical accidents. As an example, in the accident scene investigation text, there is a description of a historical accident: "The vehicle was driving on an urban elevated ramp at night in the rain. After the vehicle lost control, the driver's side door collided severely with a lamppost." The entity identified is structured data such as: "Road type = urban elevated ramp", "Cause = vehicle loss of control", "Behavior = collision", and "Weather = nighttime rainy day".

[0031] Step 1012: Use an image detection model to identify vehicle damage in the accident scene image and determine the vehicle damage feature set corresponding to the vehicle's historical accidents.

[0032] Regarding step 1012 above, in specific implementation, for the accident scene image, the accident scene image is input into a pre-trained image detection model. The image detection model is used to identify vehicle damage in the accident scene image, identifying the vehicle's deformation pattern, collision angle, and damaged parts, and determining the vehicle damage feature set corresponding to the vehicle's historical accidents. Specifically, the image detection model can output feature information including: damaged parts of the vehicle (front bumper, left A-pillar, etc.), damaged area percentage, and damage level (minor scratches, structural deformation, etc.), which are visualized through coordinate frame annotations.

[0033] Step 1013: Align the vehicle running sequence data and the accident weather data with time windows, and associate them with the accident semantic feature set and the vehicle damage feature set to obtain the target accident data.

[0034] Regarding step 1013 above, in specific implementation, the accident ID of the vehicle's historical accidents is used as the core. Structured data from different sources and extracted features are spatiotemporally aligned to form a complete accident data record. The vehicle's runtime sequence data and the weather data of the accident occurrence are aligned with a time window and associated with the accident semantic feature set and the vehicle damage feature set to obtain the target accident data. For example, the time period from 60 seconds before the accident to 10 seconds after the accident is selected, and the OBD data is aligned to the UTC timestamp at a 10Hz sampling rate, while the weather station data is interpolated to the same time granularity. The above four types of features are then cross-modal associated to generate a unified target accident data table entry, with fields including but not limited to: [Accident ID, Timestamp, Vehicle Speed ​​Sequence, Peak Acceleration, List of Damaged Parts, Weather Type, Road Type, Collision Method, Driver Behavior]. For example, the signal "Brake pedal not depressed 300ms before collision" in the EDR data is associated with the text description of "driver misoperation" in the investigation report and the information "equipped with AEB" in the vehicle configuration. In this way, the fusion strategy overcomes the problem of incomplete information from a single data source, constructing a high-fidelity data foundation that more closely approximates the full picture of a real accident. Transforming the originally chaotic and heterogeneous multi-source data into a unified, semantically rich, and structured accident database suitable for subsequent in-depth analysis is the foundation for achieving accurate scenario mining. Thus, according to steps 1011-1013 above, to achieve effective integration of multi-source data, firstly, a large language model is used to perform named entity recognition and relation extraction on the accident scene investigation text, extracting a set of accident semantic features involving key elements such as "collision object," "road type," and "driving behavior." Secondly, a pre-trained image detection model is used to locate and classify damaged areas in the accident scene images, identifying damage patterns such as front dents and side beam deformation, obtaining a set of vehicle damage features. Finally, vehicle runtime sequence data and weather data are windowed according to timestamps, and a spatiotemporal correlation is established with the aforementioned semantic features and damage features, ultimately outputting structured "target accident data." By deeply fusing multimodal heterogeneous data such as accident scene investigation text, accident images, vehicle runtime sequence data, and weather information, the problem of one-sided information from a single data source is overcome. This cross-modal fusion strategy not only improves the authenticity and completeness of accident reconstruction but also lays a high-quality data foundation for subsequent high-precision clustering analysis, solving the problem of scene modeling distortion caused by data fragmentation in traditional methods.

[0035] S102, perform cluster analysis on multiple target accident data to identify at least one risk accident scenario.

[0036] Here, cluster analysis is an unsupervised machine learning method that groups accident data with similar characteristics into the same category by calculating the similarity between samples. Risk accident scenarios are typical accident patterns with high frequency or high severity identified through clustering, such as "rear-end collisions on highways in rainy weather," representing safety hazards that require close attention.

[0037] In specific implementation of step S102, cluster analysis is performed on the multiple target accident data extracted in step S101 to identify at least one high-risk accident scenario.

[0038] As an optional embodiment, regarding step S102 above, the clustering analysis of multiple target accident data to identify at least one risk accident scenario includes: Step 1021: Convert each target accident data into a corresponding high-dimensional feature vector, and use a clustering algorithm to cluster each high-dimensional feature vector to obtain accident feature clusters with similar accident feature combinations.

[0039] Regarding step 1021 above, in a specific embodiment, each target accident data is first converted into a corresponding high-dimensional feature vector. The high-dimensional feature vector maps each accident instance to a multi-dimensional numerical array, where each dimension represents a quantifiable accident attribute, facilitating mathematical operations and pattern recognition. As an example, spatial vector conversion can be performed using an embedded model; this application does not specifically limit this approach. Then, a clustering algorithm is used to cluster each high-dimensional feature vector, resulting in accident feature clusters with similar accident feature combinations. In this way, using a clustering algorithm to group these vectors automatically discovers frequently occurring accident pattern clusters. For example, during clustering, a significant accident feature cluster might be found as: "nighttime," "rainy day," "highway," and "loss of control, skidding, and impact with guardrail." Common clustering algorithms such as k-means and DBscan can be used; this application does not specifically limit this approach. Furthermore, before clustering, the high-dimensional vectors of spatial distance can be reduced in dimensionality, for example, using PCA or t-SNE, to improve the efficiency and performance of the clustering algorithm.

[0040] Step 1022: For each accident feature cluster, the frequency of occurrence of the accident feature cluster in multiple historical accident datasets is counted, and the risk score of the accident feature cluster is determined by combining the accident impact results of the historical accidents of the corresponding vehicles.

[0041] Step 1023: When the risk score is greater than or equal to a preset score threshold, the cluster of accident features is taken as the risk accident scenario.

[0042] Regarding step 1022 above, in specific implementation, for each clustered accident feature cluster, the frequency of occurrence of that cluster in multiple historical accident datasets is statistically analyzed. The degree of accident impact can be represented by an accident impact severity factor. Specifically, it can be assessed based on the consequence information of all accidents within the cluster, including the level of personal injury (e.g., minor injury, serious injury, death), the degree of vehicle damage (repair cost or scrap rate), traffic interruption time, etc. Expert scoring or machine learning regression models can be used to normalize the multidimensional impact indicators into a severity score within the [0,1] interval. Then, the frequency of occurrence and the accident impact severity factor are weighted and summed to obtain the risk score of the accident feature cluster. If the risk score of the accident feature cluster is greater than or equal to a preset scoring threshold, the accident feature cluster is considered as the risky accident scenario.

[0043] Thus, according to steps 1021-1023 above, this application transforms each target accident data into a high-dimensional feature vector and automatically groups them into accident feature clusters using a clustering algorithm, achieving unsupervised mining of complex accident patterns. Based on this, a risk score is calculated by comprehensively considering the occurrence frequency of the cluster and its corresponding accident impact, and only when the score exceeds a threshold is it determined to be a risky accident scenario. This ensures that resources are focused on truly high-risk typical scenarios, avoids resource waste in the assessment process, and improves the scientific rigor and engineering practicality of risk identification.

[0044] S103, for each risk accident scenario, the simulation model corresponding to the target vehicle is used to simulate the risk accident scenario to evaluate the performance of the safety configuration of the target vehicle under the risk accident scenario, and determine the safety configuration evaluation result of the target vehicle.

[0045] Regarding step S103 above, in specific implementation, after identifying the risk accident scenarios, for each risk accident scenario, a simulation model corresponding to the target vehicle, including dynamics model, sensor model, control algorithm, etc., is used to simulate the risk accident scenario. The simulation observes whether the safety configuration is activated in a timely manner and effectively intervenes in the accident process, thereby evaluating the performance of the target vehicle's safety configuration under the risk accident scenario and determining the target vehicle's safety configuration evaluation result. Here, the safety configuration may include the anti-lock braking system, electronic brake force distribution system, vehicle stability control system, electronic brake assist system, etc., in the target vehicle; this application does not specifically limit this. This approach breaks through the limitations of traditional laboratory standard testing in covering limited scenarios, fully utilizes real-world accident data to mine potential high-risk conditions, and realizes a shift from an "experience-driven" to a "data-driven" safety development paradigm, significantly improving the pertinence and foresight of vehicle safety systems.

[0046] As an optional embodiment, regarding step S103 above, the step of simulating the risk accident scenario using the simulation model corresponding to the target vehicle to evaluate the performance of the target vehicle's safety configuration under the risk accident scenario and determine the safety configuration evaluation result of the target vehicle includes: Step 1031: Convert the risk incident scenario into the initial conditions corresponding to the simulation environment, and call the simulation model to run the simulation in the simulation environment to determine the intervention result of the safety configuration and the incident simulation result.

[0047] Regarding step 1031 above, in specific implementation, the key semantic features and physical parameters in the identified risk accident scenario are first mapped into numerical initial conditions that the simulation system can resolve. These initial conditions include at least one or more of the following: initial speed of the main vehicle; relative distance and relative speed (for traffic participants such as the vehicle in front and pedestrians); road geometric parameters (such as radius of curvature, slope, and lane width); road surface adhesion coefficient (derived from weather data of the accident occurrence, e.g., μ=0.4 for rainy days and μ=0.2 for icy / snowy roads); ambient light intensity and visibility (e.g., at night and in foggy weather); traffic flow density and surrounding vehicle behavior patterns; and driver reaction delay distribution (a statistical distribution derived from historical data analysis). The simulation environment can be an autonomous driving simulation platform, such as CARLA, PreScan, VTD, or the MATLAB / Simulink co-simulation framework. The simulation model corresponding to the target vehicle is a pre-constructed digital twin model, which may include a vehicle multibody dynamics model; control logic models of key safety systems, such as Automatic Emergency Braking (AEB), Electronic Stability Program (ESP), Lane Keeping Assist (LKA), etc.; sensor models (cameras, millimeter-wave radar, lidar), with the ability to model noise, occlusion, and false / missed detections; and actuator response delay models (such as brake system pressure build-up time). After setting the initial conditions, the simulation model is started and run in the simulation environment, recording the following information: whether the safety configuration is activated and when it is activated; control command output (such as deceleration requests, steering correction amounts); final accident state (such as whether a collision occurred, collision speed, peak occupant acceleration); and intermediate process variables (such as yaw rate, lateral offset).

[0048] Step 1032: Using the Monte Carlo simulation algorithm, at least one variable in the initial conditions is randomly adjusted within a preset range, and multiple repeated simulations are performed to statistically analyze the effective intervention rate and failure probability of the safety configuration under the risk accident scenario, so as to obtain the safety configuration evaluation result corresponding to the safety configuration.

[0049] Regarding step 1032 above, in order to improve the statistical representativeness and engineering applicability of the evaluation results, this embodiment introduces the Monte Carlo simulation method. Random perturbations are applied to key variables to simulate individual differences and environmental fluctuations in reality. Specifically, using the Monte Carlo simulation method, at least one variable in the initial conditions is randomly adjusted within a preset range (e.g., the main vehicle reaction delay follows a normal distribution with μ=1.2s and σ=0.3s, with uniform sampling at relative distances of [30m, 50m]). Multiple repeated simulations are performed, typically no less than 1000. Then, the effective intervention rate and failure probability of the safety configuration under this risky accident scenario are statistically analyzed to form the safety configuration evaluation results. This method, by introducing uncertainty modeling, enhances the statistical representativeness and engineering applicability of the evaluation results, outperforming the conclusions of a single deterministic simulation.

[0050] Thus, based on steps 1031-1032 above, a Monte Carlo simulation mechanism is introduced. Key variables (such as vehicle speed, reaction time, braking delay, etc.) are randomly perturbed within a preset range, and numerous repeated simulation experiments are performed to simulate uncertainties existing in reality. By statistically analyzing the performance of safety configurations under various perturbation conditions, probabilistic indicators such as "effective intervention rate" and "failure probability" are derived, which better reflect the true reliability level of the system compared to traditional single-shot deterministic simulations. This method is particularly suitable for verifying the robustness of autonomous driving or advanced driver assistance systems in edge scenarios, providing a powerful verification tool for functional safety and expected functional safety.

[0051] Specifically, regarding step 1032 above, the statistical analysis of the effective intervention rate and failure probability of the security configuration in this risk incident scenario includes: Step 10321: Analyze the simulation results of multiple accidents to determine the effective number of times the safety configuration is activated and achieves control within the preset time window, as well as the number of times the safety configuration fails to activate beyond the preset time window and fails to activate beyond the preset time window.

[0052] Regarding step 10321 above, in specific implementation, after all simulations are completed, the results of N simulations are summarized and analyzed. The effective intervention counts where the safety configuration is activated and achieves control effects within a preset time window are determined, as well as the failure counts where the safety configuration is activated beyond the preset time window and fails to activate beyond the preset time window. Here, two core judgment criteria are defined: Preset time window: refers to the time interval between when the system perceives the dangerous event and the ideal intervention time. For example, for an AEB system, if activation within 1.5 seconds before a collision is required to avoid an accident, this time is the upper limit of the effective intervention time. Effective intervention must meet two conditions: the safety configuration is activated within the preset time window; and, after activation, the expected control effect is achieved (such as significant deceleration, collision avoidance, or a reduction in collision speed ≥30%). Conversely, failures are considered, including two categories: hysteresis failure: the safety configuration activation time exceeds the preset time window; and functional deficiency failure: the safety configuration is completely not activated (e.g., due to sensor misjudgment, omission of decision logic, etc.).

[0053] Step 10322: Determine the effective intervention rate based on the number of effective interventions and the number of simulations, and determine the failure probability based on the number of failures and the number of simulations.

[0054] Regarding step 10322 above, in specific implementation, the ratio between the number of effective interventions and the number of simulations is used as the effective intervention rate, and the ratio between the number of failures and the number of simulations is used as the effective intervention rate. As an example, in a "rear-end collision on a rainy urban main road" risk scenario, after 1000 Monte Carlo simulations, the ESP system of a certain vehicle intervened too late at a specific sideslip angle, resulting in an inability to effectively avoid the accident; or its AEB system failed to identify obstacles ahead due to sensor performance degradation in rainy weather.

[0055] As an optional embodiment, after determining the safety configuration assessment results of the target vehicle, the vehicle safety assessment method provided in this application further includes: The safety configuration evaluation results are input into a pre-built vehicle engineering knowledge graph corresponding to the target vehicle to determine the safety optimization suggestions corresponding to the safety configuration.

[0056] Here, the vehicle engineering knowledge graph is a structured domain knowledge base that organizes entities in the vehicle engineering field in the form of a graph. It is constructed jointly by an expert rule base and historical R&D data. Nodes include entities such as "safety system type," "common failure modes," "influencing factors," "solutions," and "verification methods," while edges represent causal, constraint, or optimization path relationships. This knowledge graph uses "vehicle system-assembly-component" as its framework, linking a wealth of engineering knowledge. For example, it knows that "sideslip" is related to concepts such as the tuning strategy of "Electronic Stability Program (ESP)," the sensitivity of "lateral acceleration sensor," and "suspension characteristics"; it also knows that "left front collision" is related to "front longitudinal beam energy absorption structure," "engine compartment layout," and "airbag deployment sequence." Safety optimization suggestions are technical improvement recommendations derived from the knowledge graph based on the current safety configuration assessment results, such as "enhancing the target recognition capability of the AEB system under low light conditions" or "improving the thickness of the B-pillar reinforcement plate material."

[0057] Regarding the above steps, in specific implementation, after the safety configuration assessment results are determined, they are input into a pre-constructed vehicle engineering knowledge graph corresponding to the target vehicle to determine the corresponding safety optimization suggestions. For example, if the assessment results show that the AEB system's intervention failure rate on slippery roads at night reaches 25%, the system uses this as a query condition to match the subgraph structure of "low visibility + low adhesion coefficient → AEB false positives and false negatives" in the knowledge graph, and traces along the "optimization path" to feasible suggestions such as "upgrading millimeter-wave radar resolution" or "fusion with infrared camera input". This process can be automated through graph neural networks (GNN) or rule-based reasoning engines, outputting structured reports for the R&D team's reference. By introducing a knowledge graph, this application not only completes the assessment loop but also achieves an intelligent leap from problem discovery to solution proposal, greatly shortening the safety iteration cycle. Here, continuing the example above, when the system receives the analysis result that "a certain vehicle model has a high risk of sideslipping on curves in rainy weather at night, and the ESP intervention is delayed", the system will traverse the knowledge graph for safety optimization suggestions related to "ESP intervention delay", such as "it is recommended to optimize the yaw rate judgment threshold for low-adhesion road surfaces in the ESP control strategy; or upgrade the accuracy and sampling frequency of the wheel speed sensor", etc.

[0058] Thus, based on the safety configuration assessment results obtained through the above steps, this application further inputs them into a pre-constructed vehicle engineering knowledge graph. Through causal relationships and optimization path reasoning in the graph structure, executable safety optimization suggestions are automatically generated. This design breaks through the inefficient traditional process of "identifying problems—manual analysis—proposing improvements," realizing an intelligent closed loop from assessment to improvement suggestions, greatly shortening the R&D iteration cycle, and enhancing the intelligent evolution capability of the entire vehicle safety system.

[0059] Furthermore, as an example, taking the side collision battery safety risk analysis of a certain vehicle model in a specific scenario as an example, the operation process of this application is fully demonstrated: in the data aggregation and processing stage, the system accesses multiple accident reports related to the vehicle model from the database. One key report contained the following description: "A vehicle was traveling on an elevated highway ramp in the rain at night. After losing control, the driver's side door collided severely with a lamppost, and subsequently, the bottom of the vehicle caught fire." The system's natural language processing module parsed this text, accurately extracting key features such as "side pole collision," "rain," "night," "curve," and "bottom fire." Simultaneously, the system matched the vehicle's identification code to detailed configuration information, including the battery used, the battery pack's location in the chassis, and specific parameters of the side protection structure. All this information was integrated into a structured data record, laying the foundation for subsequent analysis. Entering the scene mining and simulation analysis phase, the system used an unsupervised clustering algorithm to discover that accident records with similar characteristics formed a significant high-risk cluster. Statistical analysis showed that the severity of accidents in this cluster was far higher than other types of accidents. To further investigate the root cause, the system invoked a digital twin model of the vehicle model, which accurately simulated the vehicle's body structure, material properties, and... The mechanical and thermal response of the battery pack under lateral impact was studied. Under the set simulation conditions, the system reproduced the side pole impact scenario. After multiple simulation iterations, the defects of the current design were quantified: under a side pole impact within a specific speed range, the intruding rigid pole caused excessive deformation of the B-pillar, which in turn squeezed the sidewall of the battery pack, causing the casing to tear and the cells to short-circuit, ultimately leading to thermal runaway. This finding accurately pointed out the design defect of the existing side protection structure in providing insufficient protection for the battery pack. In the safety suggestion generation stage, the system consulted the built-in automotive engineering knowledge graph, which integrates automotive engineering manuals, technical papers, and design specifications to establish a complete vehicle system knowledge system. Based on the simulation analysis results, the system automatically generated a structured safety improvement report, which detailed the identified high-risk scenarios and their root causes, and proposed two specific improvement schemes: one is to add a high-strength aluminum alloy protective beam between the side of the battery pack and the inner side of the door; the other is to upgrade the B-pillar material to higher-strength hot-formed steel in the next generation of models.

[0060] The vehicle safety assessment method based on accident data provided in this application firstly acquires a historical accident dataset corresponding to the historical accidents of the target vehicle, and then performs data fusion on the historical accident dataset to generate target accident data corresponding to the historical accidents of the vehicle; then, it performs cluster analysis on multiple target accident datasets to identify at least one risky accident scenario; finally, for each risky accident scenario, it uses a simulation model corresponding to the target vehicle to simulate the risky accident scenario in order to evaluate the performance of the safety configuration of the target vehicle under the risky accident scenario and determine the safety configuration assessment result of the target vehicle.

[0061] This application generates high-fidelity target accident data by acquiring historical accident datasets of target vehicles and fusing multi-source data. It then performs cluster analysis on multiple target accident datasets to automatically identify high-frequency or high-hazard risk accident scenarios. Finally, it uses simulation models to evaluate the performance of vehicle safety configurations under these risk scenarios, ultimately outputting quantitative evaluation results. Compared to traditional safety assessment methods that rely on fixed standard tests, this application proactively identifies potential high-risk scenarios not covered by standard tests through in-depth mining of massive, multi-dimensional accident data. These identified high-risk scenarios are then simulated and tested in vehicle digital models, quantifying the effectiveness boundaries of safety configurations. This allows for the evaluation of vehicle safety configurations under extreme conditions before large-scale real-world accidents occur, significantly improving the relevance, foresight, and real-world coverage of safety assessments.

[0062] Please see Figure 2 , Figure 3 , Figure 2 This is one of the structural schematic diagrams of a vehicle safety assessment device based on accident data provided in an embodiment of this application. Figure 3 This is a second schematic diagram of a vehicle safety assessment device based on accident data provided in an embodiment of this application. Figure 2 As shown, the vehicle safety assessment device 200 includes: The accident data acquisition module 201 is used to acquire the historical accident dataset corresponding to the historical accidents of the target vehicle, and to perform data fusion on the historical accident dataset to generate the target accident data corresponding to the historical accidents of the vehicle. The risk accident scenario determination module 202 is used to perform cluster analysis on multiple target accident data to identify at least one risk accident scenario. The safety assessment module 203 is used to simulate each risk accident scenario using the simulation model corresponding to the target vehicle, in order to evaluate the performance of the safety configuration of the target vehicle under the risk accident scenario and determine the safety configuration assessment result of the target vehicle.

[0063] Furthermore, such as Figure 3 As shown, the vehicle safety assessment device 200 further includes an optimization suggestion determination module 204. After determining the safety configuration assessment result of the target vehicle, the optimization suggestion determination module 204 is used to: The safety configuration evaluation results are input into a pre-built vehicle engineering knowledge graph corresponding to the target vehicle to determine the safety optimization suggestions corresponding to the safety configuration.

[0064] Furthermore, the historical accident dataset includes accident scene investigation text, accident scene images, vehicle operation sequence data, and weather data of the accident occurrence; the accident data acquisition module 201 is used to perform data fusion on the historical accident dataset to generate target accident data corresponding to the vehicle's historical accidents, including: Using a large language model, named entity recognition and relation extraction are performed on the accident scene investigation text to determine the accident semantic feature set corresponding to the vehicle's historical accidents; The image detection model is used to identify vehicle damage in the accident scene images to determine the vehicle damage feature set corresponding to the vehicle's historical accidents. The vehicle runtime sequence data and the accident weather data are aligned with a time window and associated with the accident semantic feature set and the vehicle damage feature set to obtain the target accident data.

[0065] Furthermore, when the risk accident scenario determination module 202 is used to perform cluster analysis on multiple target accident data to identify at least one risk accident scenario, the risk accident scenario determination module 202 is also used to: Each target accident data is converted into a corresponding high-dimensional feature vector, and a clustering algorithm is used to cluster each high-dimensional feature vector to obtain accident feature clusters with similar accident feature combinations. For each accident feature cluster, the frequency of occurrence of the accident feature cluster in multiple historical accident datasets is counted, and the risk score of the accident feature cluster is determined by combining the accident impact results of the historical accidents of the corresponding vehicles. When the risk score is greater than or equal to a preset score threshold, the cluster of accident features is taken as the risk accident scenario.

[0066] Furthermore, when the safety assessment module 203 is used to simulate the risk accident scenario using the simulation model corresponding to the target vehicle to evaluate the performance of the target vehicle's safety configuration under the risk accident scenario and determine the safety configuration assessment result of the target vehicle, the safety assessment module 203 is also used to: The risk incident scenario is converted into the initial conditions corresponding to the simulation environment, and the simulation model is called to run the simulation in the simulation environment to determine the intervention result of the safety configuration and the incident simulation result; Using the Monte Carlo simulation algorithm, at least one variable in the initial conditions is randomly adjusted within a preset range, and multiple repeated simulations are performed. The effective intervention rate and failure probability of the safety configuration under the risk incident scenario are statistically analyzed to obtain the safety configuration evaluation result corresponding to the safety configuration.

[0067] Furthermore, when the security assessment module 203 is used to statistically analyze the effective intervention rate and failure probability of the security configuration in the risk incident scenario, the security assessment module 203 is also used to: The results of multiple accident simulations were analyzed to determine the effective number of times the safety configuration was activated and achieved control within a preset time window, as well as the number of times the safety configuration failed to activate beyond the preset time window and failed to activate beyond the preset time window. The effective intervention rate is determined based on the number of effective interventions and the number of simulations, and the failure probability is determined based on the number of failures and the number of simulations.

[0068] Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 4 As shown, the electronic device 400 includes a processor 410, a memory 420, and a bus 430.

[0069] The memory 420 stores machine-readable instructions executable by the processor 410. When the electronic device 400 is running, the processor 410 communicates with the memory 420 via the bus 430. When the machine-readable instructions are executed by the processor 410, they can perform the operations described above. Figure 1 The specific implementation of the vehicle safety assessment method based on accident data in the method embodiment shown can be found in the method embodiment, and will not be repeated here.

[0070] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can perform the above-described actions. Figure 1 The specific implementation of the vehicle safety assessment method based on accident data in the method embodiment shown can be found in the method embodiment, and will not be repeated here.

[0071] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0072] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the shown or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0073] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0074] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0075] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0076] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The scope of protection of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A vehicle safety assessment method based on accident data, characterized in that, The vehicle safety assessment method includes: Obtain the historical accident dataset corresponding to the historical accidents of the target vehicle, and perform data fusion on the historical accident dataset to generate the target accident data corresponding to the historical accidents of the vehicle. Cluster analysis was performed on multiple target accident data to identify at least one risk accident scenario; For each risk incident scenario, the simulation model corresponding to the target vehicle is used to simulate the risk incident scenario in order to evaluate the performance of the safety configuration of the target vehicle under the risk incident scenario and determine the safety configuration evaluation result of the target vehicle.

2. The vehicle safety assessment method according to claim 1, characterized in that, After determining the safety configuration assessment results of the target vehicle, the vehicle safety assessment method further includes: The safety configuration evaluation results are input into a pre-built vehicle engineering knowledge graph corresponding to the target vehicle to determine the safety optimization suggestions corresponding to the safety configuration.

3. The vehicle safety assessment method according to claim 1, characterized in that, The historical accident dataset includes accident scene investigation text, accident scene images, vehicle operation sequence data, and weather data of the accident occurrence; The historical accident dataset is fused to generate target accident data corresponding to the vehicle's historical accidents, including: Using a large language model, named entity recognition and relation extraction are performed on the accident scene investigation text to determine the accident semantic feature set corresponding to the vehicle's historical accidents; The image detection model is used to identify vehicle damage in the accident scene images to determine the vehicle damage feature set corresponding to the vehicle's historical accidents. The vehicle runtime sequence data and the accident weather data are aligned with a time window and associated with the accident semantic feature set and the vehicle damage feature set to obtain the target accident data.

4. The vehicle safety assessment method according to claim 1, characterized in that, The clustering analysis of multiple target accident data to identify at least one risky accident scenario includes: Each target accident data is converted into a corresponding high-dimensional feature vector, and a clustering algorithm is used to cluster each high-dimensional feature vector to obtain accident feature clusters with similar accident feature combinations. For each accident feature cluster, the frequency of occurrence of the accident feature cluster in multiple historical accident datasets is counted, and the risk score of the accident feature cluster is determined by combining the accident impact results of the historical accidents of the corresponding vehicles. When the risk score is greater than or equal to a preset score threshold, the cluster of accident features is taken as the risk accident scenario.

5. The vehicle safety assessment method according to claim 1, characterized in that, The step of simulating the risky accident scenario using a simulation model corresponding to the target vehicle to evaluate the performance of the target vehicle's safety configuration under the risky accident scenario and determine the safety configuration evaluation result of the target vehicle includes: The risk incident scenario is converted into the initial conditions corresponding to the simulation environment, and the simulation model is called to run the simulation in the simulation environment to determine the intervention result of the safety configuration and the incident simulation result; Using the Monte Carlo simulation algorithm, at least one variable in the initial conditions is randomly adjusted within a preset range, and multiple repeated simulations are performed. The effective intervention rate and failure probability of the safety configuration under the risk incident scenario are statistically analyzed to obtain the safety configuration evaluation result corresponding to the safety configuration.

6. The vehicle safety assessment method according to claim 5, characterized in that, The statistics on the effective intervention rate and failure probability of the security configuration in this risk incident scenario include: The results of multiple accident simulations were analyzed to determine the effective number of times the safety configuration was activated and achieved control within a preset time window, as well as the number of times the safety configuration failed to activate beyond the preset time window and failed to activate beyond the preset time window. The effective intervention rate is determined based on the number of effective interventions and the number of simulations, and the failure probability is determined based on the number of failures and the number of simulations.

7. A vehicle safety assessment device based on accident data, characterized in that, The vehicle safety assessment device includes: The accident data acquisition module is used to acquire the historical accident dataset corresponding to the historical accidents of the target vehicle, and to perform data fusion on the historical accident dataset to generate the target accident data corresponding to the historical accidents of the vehicle. The risk incident scenario identification module is used to perform cluster analysis on multiple target incident data to identify at least one risk incident scenario. The safety assessment module is used to simulate each risk accident scenario using the simulation model corresponding to the target vehicle, in order to evaluate the performance of the safety configuration of the target vehicle under the risk accident scenario and determine the safety configuration assessment result of the target vehicle.

8. The vehicle safety assessment device according to claim 7, characterized in that, The vehicle safety assessment device further includes an optimization suggestion determination module. After determining the safety configuration assessment result of the target vehicle, the optimization suggestion determination module is used to: The safety configuration evaluation results are input into a pre-built vehicle engineering knowledge graph corresponding to the target vehicle to determine the safety optimization suggestions corresponding to the safety configuration.

9. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. The machine-readable instructions are executed by the processor to perform the steps of the vehicle safety assessment method based on accident data as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the vehicle safety assessment method based on accident data as described in any one of claims 1 to 6.

Images