Automatic driving vehicle accident cause analysis method

Abstract

The invention discloses an automatic driving vehicle accident cause analysis method. The method comprises the steps of obtaining a data set of an automatic driving vehicle accident report; pre-processing the data set, wherein the pre-processing comprises abnormal value cleaning, error data correction and missing data repair; constructing an explanation variable system of an automatic driving vehicle accident, and carrying out statistical description on accident characteristics; an accident cause analysis model is provided by integrating existing automatic driving vehicle accident research data and conclusions based on a random parameter Logit method for solving a heterogeneity problem; based on priori knowledge and an analysis result, designing a random parametric variable, and constructing an RPL model: sampling a probability density function for multiple times by adopting a Monte Carlo method, and taking a simulated probability mean value as an integral approximate solution; and the analysis of the variable random parameter effect in the automatic driving vehicle accident is realized through the RPL model. According to the invention, the cause explanation of the heterogeneity of the important characteristics and influence factors of the accident of the automatic driving vehicle is realized.

Description

Technical Field

[0001] The present invention relates to the field of analysis of causes of traffic accidents, and particularly to a method for analyzing causes of accidents of autonomous vehicles. Background Art

[0002] At present, autonomous vehicle traffic accidents have received extensive attention. However, due to the difficulties in collecting traffic accident data and the scarcity of data, the analysis of autonomous vehicle traffic accidents is relatively limited. The research on autonomous vehicle accidents focuses on accident prevention technologies, including collision avoidance, tracking lane positions, and enhancing V2X communication. At present, the core technologies of autonomous vehicles are showing a diversified development trend, and there is a lack of post-accident evaluation and feedback on core technologies such as different perception devices, target detection algorithms, obstacle avoidance mechanisms, and human-machine takeover solutions. There is a problem that the mining and analysis of autonomous vehicle accident data are not deep enough.

[0003] The reconstruction of explanatory variables for autonomous vehicle accidents can better describe the occurrence of accidents and reveal the important factors affecting accidents. The interpretation of accident mechanisms depends on the excavation of differential autonomous driving technology variables, the analysis of accident characteristics, the determination of key accident objects, the correct construction of causal models, and the interpretation of their heterogeneity.

[0004] Traditional traffic accident analysis elements usually include four elements: drivers, vehicles, roads, and environments. However, autonomous vehicle accidents have important influencing factors related to autonomous driving technologies such as autonomous driving levels, perception types, decision-making and execution situations, etc. These influencing factors are complex and diverse, and there is a lack of corresponding post-accident evaluation and feedback.

[0005] There are differences in autonomous vehicle accidents. Compared with conventional vehicles, autonomous vehicles are more prone to side collision accidents and rear-end collision accidents; among intelligent connected vehicles, vehicles equipped with ADAS autonomous driving levels have a higher risk of frontal collision, and vehicles with ADS autonomous driving levels have a higher risk of rear-end collision.

[0006] The specific influencing variables and their action mechanisms under different collision types are also different. The cause analysis of autonomous vehicle accidents needs to pay attention to the differences in collision positions and accident severity differences between them and traditional accidents, and solve the important problems of autonomous vehicles occurring in multi-category collision accidents (frontal collision, side collision, rear-end collision) and accident severity (injured, uninjured).

[0007] This application can expand the breadth and depth of influencing factors of autonomous vehicle accidents, provide a research direction and lay a theoretical foundation for the cause analysis of autonomous vehicle accidents, and provide support for the improvement of the safety performance of autonomous vehicles and the formulation of laws and regulations. Summary of the Invention

[0008] In view of the defects existing in the prior art solutions, on the first hand, the present invention provides a method for analyzing the causes of autonomous vehicle accidents, including the following steps: S1. Obtain a data set of autonomous vehicle accident reports; S2. Preprocess the data set, where the preprocessing includes outlier cleaning, error data correction, and missing data patching; S3. Construct an explanatory variable system for autonomous vehicle accidents and conduct statistical descriptions of accident characteristics; S4. Based on the existing research data and conclusions of autonomous vehicle accidents and the random parameter Logit method for solving the heterogeneity problem, propose an accident cause analysis model; S5. Based on prior knowledge and analysis results, design random parameter variables and construct an RPL model: ; ; wherein, x i is the set of accident explanatory variables, β is the independent variable x i is the parameter set of J is the set of accident severities, including injured and uninjured; β ijk is β the i th j th component of the k th dependent variable of the μ jk is the parameter mean of the j th independent variable of the k th dependent variable, σ jk is the parameter standard deviation of the j th independent variable of the k th dependent variable, υ ijk is the unobserved random effect of the i th independent variable of the j th dependent variable of the k th accident sample; S6. Adopt the Monte Carlo method to conduct multiple samplings on the probability density function to simulate the probability mean as the integral approximate solution: ; ; wherein, L is the log-likelihood function, N is the sample size of the accident data, ωij is a 0 / 1 variable. When the dependent variable is j 1 when it is, otherwise 0; P ij is the simulation probability, R The total number of samples set; S7. Analyze the random parameter effects of variables in autonomous driving vehicle accidents through the RPL model.

[0009] Furthermore, based on the data set of accident reports of autonomous vehicles obtained, the collision type and severity are combined, including frontal collision injury, frontal collision no injury, side collision injury, side collision no injury, rear-end collision injury, and rear-end collision no injury.

[0010] Furthermore, the data set of the autonomous driving vehicle accident report includes autonomous driving technology factors, vehicle factors, road factors, and environmental factors; the autonomous driving technology factors include decision-execution status, perception type, autonomous driving level, autonomous driving vehicle pre-collision motion, and model year; the vehicle factors include operation type, conventional vehicle pre-collision motion, and airbags; the road factors include road surface, road conditions, and road type; and the environmental factors include season, lighting, weather, and collision object.

[0011] Furthermore, an explanatory variable system for the autonomous driving vehicle accidents is constructed from the perspectives of the four elements of vehicle, road, environment and autonomous driving core technology.

[0012] Furthermore, S5 specifically includes taking the joint result of the collision type and severity of the autonomous driving vehicle accident as the response variable, based on the explanatory variable system of the autonomous driving vehicle accident, taking the autonomous driving technology as the key factor affecting the accident, using simulation to solve the RPL model, and obtaining the calibration results of the model parameters.

[0013] Furthermore, in S5, the calibration results of the random parameters and effect equations of the RPL model are: ; ; ; ; ; ; ; ; in, UFC No injuries in a forward collision. ILC Injured in a side collision.ULC Not injured in a side collision, IRC Injured in a rear-end collision, URC Not injured in a rear-end collision; x 1 ~x 13 There are 13 independent variables affecting the accident, corresponding to the ADS autonomous driving level, single sensor, multi-sensor, radical decision-making - execution situation, autonomous vehicle going straight, autonomous vehicle turning, driverless, conventional vehicle turning, conventional vehicle parking, non-motor vehicle, highway, intersection, and road obstacles respectively.

[0014] Further, the processing language of the dataset is the Python programming language.

[0015] Further, the S7 specifically includes: based on the effect equation and random parameters of RPL, inputting the explanatory variable system and calibrating the parameters using the simulation method, screening out insignificant variables using the stepwise regression method until all variables are significant after iteration, solving the marginal effects of each variable and outputting, to complete the accident causation analysis method based on the RPL model.

[0016] In a second aspect, the present invention also provides an electronic device, including a memory and a processor, where the memory is used to store one or more computer instructions, and wherein the one or more computer instructions are executed by the processor to implement an accident causation analysis method for autonomous vehicles as described above.

[0017] In a third aspect, a readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the steps of an accident causation analysis method for autonomous vehicles as described above.

[0018] Advantages of the present invention: 1. The present invention provides an accident causation analysis method for autonomous vehicles. By adding autonomous driving core technology variables to the accident explanatory variable system to represent autonomous vehicle accidents, based on designing random parameter variables, constructing an RPL model with the accident collision type and severity joint variables as response variables, and taking the publicly available autonomous vehicle accident dataset as an example to conduct accident causation analysis for autonomous vehicles.

[0019] 2. The present invention solves the problem that under the background of the development of differentiated autonomous driving technologies, the evaluation of the risk avoidance effect and the elaboration of the action mechanism of the core technologies in autonomous vehicle accidents have not been clarified, and realizes the causal explanation of the important characteristics and heterogeneity of influencing factors in autonomous vehicle accidents.

[0020] ​3. The present invention is based on setting random parameters, designs a group of effect equations of random parameter Logit, combines the important characteristics and influencing factors of autonomous driving vehicle accidents to form an accident cause analysis method, and extracts variables based on public autonomous driving vehicle accident reports to analyze the characteristics and causal mechanisms of autonomous driving vehicle accidents.

[0021] 4. This invention expands the explanatory variable system of autonomous driving vehicle accidents, analyzes the impact mechanism of differentiated core technologies, and can provide support for the optimization of autonomous driving vehicles and the formulation of relevant laws and regulations. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The present invention has the following accompanying drawings: Figure 1 is a flow chart of the steps of the present invention; Figure 2 is a schematic diagram of the accident explanation variable system of the autonomous driving vehicle of the present invention; Figure 3 It is a schematic diagram of the fitting process of the RPL causal analysis model of the present invention; Figure 4 It is a distribution diagram of the heterogeneity parameters of the RPL causal analysis model of the present invention. DETAILED DESCRIPTION

[0023] To make the objects, advantages and features of the present invention more apparent, the following Figures 1-4 The present invention is further described in detail with reference to the following specific embodiments.

[0024] Embodiment 1: like Figure 1 As shown, the present invention provides a method for analyzing the cause of an accident of an autonomous driving vehicle, which mainly includes the following steps: Step 1: Obtain a dataset of accident reports of autonomous vehicles from the official websites of relevant departments, and combine the collision type and severity according to actual research needs, i.e., frontal collision with injury, frontal collision without injury, side collision with injury, side collision without injury, rear-end collision with injury, and rear-end collision without injury. Obtain the core technology information of autonomous driving of the research vehicle from the websites of relevant companies, including autonomous driving level, perception type, decision execution, model year, etc., and match it with the model in the dataset.

[0025] Step 2: Preprocess the acquired multi-category autonomous driving vehicle accident data, including outlier cleaning, erroneous data correction, missing data repair, etc.

[0026] Step 3: Construct an explanatory variable system for autonomous driving vehicle accidents from the perspectives of autonomous driving technology, vehicle, road, and environment, and perform a statistical description of accident characteristics. Figure 2As shown, the autonomous driving technical factors include decision - execution situation, perception type, autonomous driving level, pre - collision movement of autonomous vehicles, and model year; the vehicle factors include operation type, pre - collision movement of conventional vehicles, and airbag; the road factors include road surface, road condition, and road type; the environmental factors include season, lighting, weather, and collision object.

[0027] Step 4: Consider the random parameter effects existing in the influence of variables and the correlation between collision type and severity, such as Figure 3 As shown, based on prior knowledge, formulate the effect equation and random parameters of the RPL, input the explanatory variable system constructed in Step 3, and use the simulation method to calibrate the parameters. Use the stepwise regression method to screen out insignificant variables until all variables are significant after iteration, solve the marginal effects of each variable and output, thus completing the accident causation analysis method based on the RPL model.

[0028] Step 5: Take the combined result of the collision type and severity of autonomous vehicle accidents as the response variable, take the explanatory variables of autonomous vehicle accidents as the independent variables, regard autonomous driving technology as the key factor of concern in the accident, and use the simulation method to solve the RPL model to obtain the calibration result of the model parameters. The solution of the RPL model needs to be implemented using NLogit software.

[0029] Based on the calibration result of the model variable parameters and the marginal effects, analyze the influence of significant variables to achieve the goals of analyzing the causation mechanism of multi - category autonomous vehicle accidents and post - accident assessment of autonomous driving technology.

[0030] The accident data described in Step 1 requires a certain degree of accuracy, and autonomous vehicle models need to be matched according to their core technology types. The core obstacle avoidance mechanisms of autonomous driving for different models are different.

[0031] Steps 2 and 3 need to use programming languages such as Python for data processing.

[0032] The index system established in Step 3 can be changed according to actual research needs.

[0033] In Step 5, based on prior knowledge and analysis results, design random parameter variables, and the established RPL model is: ; ; Where x i is the set of accident explanatory variables, β is the independent variable x i is the set of parameters of J is the set of accident severity (divided into injured and uninjured);β ijk is β the i th j component of the k th μ jk is the parameter mean of the j th k independent variable of the σ jk is the parameter standard deviation of the j th k independent variable of the υ ijk is the unobserved random effect of the i th j dependent variable of the k independent variable for the

[0034] The calibration results of the random parameters and effect equations of the RPL model in Step 5 are as follows: ; ; ; ; ; ; ; ; where UFC is not injured in a frontal collision, ILC is injured in a side collision, ULC is not injured in a side collision, IRC is injured in a rear-end collision, URC is not injured in a rear-end collision; x 1 ~x 13 are 13 dependent variables affecting accidents, corresponding respectively to the ADS autopilot level, single sensor, multi-sensor, aggressive decision-making - execution, straight driving of autonomous vehicles, turning of autonomous vehicles, driverless, turning of conventional vehicles, parking of conventional vehicles, non-motor vehicles, highways, intersections, and road obstacles.

[0035] The RPL model in Step 5 shows that turning of conventional vehicles, driverless, and the ADS autopilot level have heterogeneous effects on not being injured in a rear-end collision. The heterogeneous effect situation is as Figure 4As shown, among them, the heterogeneity of the turning performance of conventional vehicles follows a normal distribution of N(-1.269, 1.5002); the driverless / remote testing of autonomous vehicles follows a normal distribution of N(-3.469, 5.0322); the ADS system follows a normal distribution of N(2.507, 3.6872) in the case of rear-end collisions without injury.

[0036] The random effect parameters of the RPL model in step 5 show that in a small number of cases, the turning of conventional vehicles increases the risk of injury in rear-end collisions. These cases may be that the complex behavior of conventional vehicles affects the decision-making of autonomous vehicles, or other types of collisions evolve into rear-end collisions. There are a small number of cases where driverless operation reduces the risk of injury. There may be passengers or safety officers inside the autonomous vehicle in the driverless mode. Therefore, the operation type of driverless / remote testing may lead to very different accident injury situations. The risk avoidance ability of a small part of the ADS system is not good. On the one hand, the vehicle type / manufacturer is an important factor affecting autonomous vehicle accidents; on the other hand, for the same type of autonomous vehicle, when it faces a small number of complex road conditions, there may be different performances due to differences in system reinforcement learning.

[0037] The RPL model in step 5 realizes the analysis of the random parameter effects of variables in autonomous vehicle accidents. By analyzing the results of the model, the influencing variables of the severity under various collision types can be explained.

[0038] In the present invention, taking the publicly available autonomous vehicle accident dataset as an example, an analysis model of autonomous vehicle accidents from July 2021 to August 2024 is constructed and the mechanism of action of the influencing factors is explained.

[0039] Embodiment 2: An electronic device includes a memory and a processor. The memory is used to store one or more computer instructions, and one or more computer instructions are executed by the processor to implement the above method of interfering with a network scanner.

[0040] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working process of the above-described electronic device can refer to the corresponding process in the foregoing method embodiment, and will not be described in detail here.

[0041] Embodiment 3: A computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, the steps of the method in Embodiment 1 are implemented.

[0042] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a device, or a computer program product. Therefore, the present invention can take the form of an all-hardware embodiment, an all-software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) that contain computer-usable program code.

[0043] The present invention is described with reference to the flowcharts and / or block diagrams of methods, terminal devices (systems), and computer program products according to the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal devices generate a device for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0044] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal devices to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0045] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal devices, so that a series of operation steps are executed on the computer or other programmable terminal devices to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable terminal devices provide steps for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0046] The present invention not only reconstructs the explanatory variable system of autonomous vehicle accidents, enabling it to better characterize accident characteristics, but also can refine the causal mechanisms of multiple categories of accident types, analyze the influence of random parameter effect variables, and lay a theoretical foundation and provide technical support for improving the safety of autonomous vehicles.

[0047] The specific embodiments described above have further elaborated on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above description is only for the specific embodiments of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for analyzing the causes of accidents in autonomous vehicles, characterized in that, The steps include: S1. Obtain a dataset of autonomous vehicle accident reports; S2, preprocessing the data set, wherein the preprocessing includes outlier cleaning, erroneous data correction, and missing data repair; S3. Construct an explanatory variable system for autonomous driving vehicle accidents and provide a statistical description of accident characteristics; S4. Based on the existing research data and conclusions on autonomous driving vehicle accidents, an accident cause analysis model is proposed based on the random parameter Logit method to solve the heterogeneity problem; S5. Based on prior knowledge and analysis results, design random parameter variables and build the RPL model: ； ； Among them, x i is the set of accident explanatory variables, β is the independent variable x i 's parameter set, J is the set of accident severity, including injured and uninjured; β ijk is β the i th component of the j th dependent variable of the k th accident sample; μ jk is the parameter mean of the j th dependent variable and the k th independent variable, σ jk is the parameter standard deviation of the j th dependent variable and the k th independent variable, υ ijk is the unobserved random effect of the i th accident sample, the j th dependent variable, and the k th independent variable; S6. Use the Monte Carlo method to sample the probability density function multiple times and use the simulated probability mean as the integral approximate solution: ; ; Among them, L is the log-likelihood function, N is the sample size of accident data, ω ij is a 0 / 1 variable, which is 1 when the dependent variable is j and 0 otherwise; P ij is the simulation probability, R is the total number of samplings set; S7. Analyze the random parameter effects of variables in autonomous driving vehicle accidents through the RPL model.

2. The method for analyzing the causes of an autonomous vehicle accident according to claim 1, characterized in that, According to the dataset of accident reports of autonomous vehicles, the collision type and severity are combined, including frontal collision injury, frontal collision no injury, side collision injury, side collision no injury, rear-end collision injury, and rear-end collision no injury.

3. The method for analyzing the causes of an autonomous vehicle accident according to claim 2, wherein The data set of the autonomous driving vehicle accident report includes autonomous driving technology factors, vehicle factors, road factors, and environmental factors; the autonomous driving technology factors include decision-execution status, perception type, autonomous driving level, autonomous driving vehicle pre-collision motion, and model year; the vehicle factors include operation type, conventional vehicle pre-collision motion, and airbags; the road factors include road surface, road conditions, and road type; the environmental factors include season, lighting, weather, and collision object.

4. The method for analyzing the causes of an autonomous vehicle accident according to claim 3, characterized in that, The explanatory variable system of the autonomous driving vehicle accidents is constructed from the four elements of vehicle, road, environment and autonomous driving core technology.

5. The method for analyzing the causes of an autonomous vehicle accident according to claim 4, wherein, The S5 specifically includes taking the joint result of the collision type and severity of the autonomous driving vehicle accident as the response variable, based on the autonomous driving vehicle accident explanatory variable system, taking the autonomous driving technology as the key factor affecting the accident, using simulation to solve the RPL model, and obtaining the calibration results of the model parameters.

6. The method for analyzing the causes of an autonomous vehicle accident according to claim 5, wherein In S5, the calibration results of the random parameters and effect equations of the RPL model are: ； ； ； ； ； ； ； ； Among them, UFC is not injured in a frontal collision, ILC is injured in a side collision, ULC is not injured in a side collision, IRC is injured in a rear-end collision, URC is not injured in a rear-end collision; x 1 ~x 13 are 13 independent variables affecting the accident, corresponding to the ADS autonomous driving level, single sensor, multi-sensor, radical decision-making - execution situation, autonomous vehicle going straight, autonomous vehicle turning, driverless, conventional vehicle turning, conventional vehicle parking, non-motor vehicle, highway, intersection, and road obstacle respectively.

7. The method for analyzing the cause of an accident of an autonomous vehicle according to claim 6, characterized in that, The data set is processed in the Python programming language.

8. The method for analyzing the causes of an autonomous vehicle accident according to claim 7, wherein, The S7 specifically includes, based on the effect equation and random parameters of RPL, inputting the explanatory variable system and calibrating the parameters using the simulation method, applying the stepwise regression method to screen out insignificant variables until all variables are significant through iteration, solving the marginal effect of each variable and outputting it, and completing the accident cause analysis method based on the RPL model.

9. An electronic device, characterized in that, It includes a memory and a processor, the memory is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement an automatic driving vehicle accident cause analysis method as described in any one of claims 1 to 8.

10. A readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of an accident cause analysis method for an autonomous driving vehicle as described in any one of claims 1 to 8 are implemented.

Images