Driving risk assessment method and device, electronic equipment and storage medium

Abstract

The invention provides a driving risk assessment method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining driving risk source information and a driving risk result of a vehicle, inputting a traffic participant aggregation behavior into a pre-trained prediction model, outputting a first prediction result of the influence of the traffic participant aggregation behavior on the driving risk result, and inputting the risk confounding factor into a pre-trained interference term model, outputting a second prediction result that the risk confounding factor affects the aggregation behavior of the traffic participants, calculating a residual error between the driving risk result and the sum of the first prediction result and the second prediction result, and obtaining an average causal effect that the aggregation behavior of the traffic participants affects the driving risk result. According to the method, the risk hybrid factors in the driving risk source are eliminated firstly, then the average causal effect of the traffic participant gathering behavior on the driving risk result is calculated, the causal relationship between the traffic participant and the driving risk is accurately analyzed, and the accuracy of driving risk cause analysis is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of autonomous driving, and particularly to a driving risk assessment method, device, electronic device, and storage medium. Background Art

[0002] With the rapid development of autonomous driving technology, autonomous vehicles are increasingly widely used on the road. Among them, the safety of the autonomous driving system has always been the focus of user attention. Accurately analyzing the causes of autonomous driving risks is crucial for improving the safety and robustness of autonomous driving.

[0003] Currently, in the autonomous driving scenario, bad weather is a typical risk confounding factor. Bad weather may simultaneously affect the behaviors of traffic participants, thereby having a complex impact on driving risks. However, existing driving risk cause analysis is usually based on fixed causal relationship assumptions and data statistical models, without considering the interference of risk confounding factors. Directly performing autonomous driving risk cause analysis on the behaviors of traffic participants will result in a large error in the driving risk cause analysis, further affecting the accuracy of the model in analyzing the causal relationship between traffic participants and driving risks. Summary of the Invention

[0004] In view of this, the present invention aims to provide a driving risk assessment method, device, electronic device, and storage medium to solve the problem that the current direct analysis of the causes of autonomous driving risks for the behaviors of traffic participants without considering the interference of risk confounding factors leads to a large error in the driving risk cause analysis.

[0005] According to the first aspect of the present invention, a driving risk assessment method is provided. The method includes:

[0006] Obtain the driving risk source information and driving risk results of the vehicle, where the driving risk source information includes traffic participant aggregation behaviors and risk confounding factors;

[0007] Input the traffic participant aggregation behaviors into a pre-trained prediction model, and output a first prediction result of the influence of the traffic participant aggregation behaviors on the driving risk results;

[0008] Input the risk confounding factors into a pre-trained interference term model, and output a second prediction result of the influence of the risk confounding factors on the traffic participant aggregation behaviors;

[0009] Calculate the residual between the driving risk results and the sum of the first prediction result and the second prediction result to obtain the average causal effect of the traffic participant aggregation behaviors on the driving risk results.

[0010] Optionally, obtain the driving risk source information and driving risk results of the vehicle, where the driving risk source information includes traffic participant aggregation behavior and risk confounding factors, including:

[0011] When it is determined that there is a driving risk during vehicle operation, obtain the driving risk result, as well as the traffic participant information and risk confounding factors corresponding to the driving risk result;

[0012] Calculate the aggregation distance in the traffic participant information, and merge the traffic participant information with an aggregation distance less than the preset distance into traffic participant aggregation behavior;

[0013] Determine the traffic participant aggregation behavior and the risk confounding factors as the driving risk source information.

[0014] Optionally, input the traffic participant aggregation behavior into a pre-trained prediction model, and output a first prediction result of the influence of the traffic participant aggregation behavior on the driving risk result, including:

[0015] Use multiple decision trees in the prediction model to predict the traffic participant aggregation behavior, and obtain multiple prediction results of the traffic participant aggregation behavior;

[0016] Perform mean processing on the multiple prediction results of the traffic participant aggregation behavior, and determine the mean prediction result as the first prediction result of the influence of the traffic participant aggregation behavior on the driving risk result.

[0017] Optionally, input the risk confounding factors into a pre-trained interference term model, and output a second prediction result of the influence of the risk confounding factors on the traffic participant aggregation behavior, including:

[0018] Use multiple decision trees in the interference term model to predict the risk confounding factors, and obtain multiple prediction results of the risk confounding factors;

[0019] Perform mean processing on the multiple prediction results of the risk confounding factors, and determine the mean prediction result as the second prediction result of the influence of the risk confounding factors on the traffic participant aggregation behavior.

[0020] Optionally, after obtaining the driving risk source information and driving risk results of the vehicle, where the driving risk source information includes traffic participant aggregation behavior and risk confounding factors, it further includes;

[0021] Randomly sample the traffic participant aggregation behavior, the risk confounding factors, and the driving risk results to obtain a sampling sample set;

[0022] Train a pre-constructed decision tree using the sampled sample set to obtain a prediction model and a nuisance model;

[0023] According to the type of quantity to be predicted, call the prediction model and / or the nuisance model.

[0024] Optionally, calculate the residual of the driving risk result minus the sum of the first prediction result and the second prediction result to obtain the average causal effect of the traffic participant aggregation behavior on the driving risk result. The average causal effect of the traffic participant aggregation behavior on the driving risk result is calculated by the following formula:

[0025] ATE = Y c - g _pred - f _pred

[0026]

[0027] where ATE is the average causal effect of the traffic participant aggregation behavior on the driving risk result, Y c is the driving risk result, g _pred is the first prediction result of the traffic participant aggregation behavior on the driving risk result, f _pred is the second prediction result of the risk confounding factor on the traffic participant aggregation behavior, T n (x) is the traffic participant aggregation behavior, W n (x) is the risk confounding factor, and N is the number of decision trees.

[0028] According to a second aspect of the present invention, there is provided a driving risk assessment device, the device comprising:

[0029] An acquisition module, configured to acquire driving risk source information and a driving risk result of a vehicle, wherein the driving risk source information includes traffic participant aggregation behavior and a risk confounding factor;

[0030] A first processing module, configured to input the traffic participant aggregation behavior into a pre-trained prediction model, and output a first prediction result of the traffic participant aggregation behavior on the driving risk result;

[0031] A second processing module, configured to input the risk confounding factor into a pre-trained nuisance model, and output a second prediction result of the risk confounding factor on the traffic participant aggregation behavior;

[0032] An average causal effect module, configured to calculate the residual of the driving risk result minus the sum of the first prediction result and the second prediction result, to obtain the average causal effect of the traffic participant aggregation behavior on the driving risk result.

[0033] Optionally, the obtaining module includes:

[0034] An obtaining sub-module, configured to obtain a driving risk result, traffic participant information corresponding to the driving risk result, and a risk confounding factor when it is determined that there is a driving risk during vehicle operation;

[0035] An aggregation sub-module, configured to calculate an aggregation distance in the traffic participant information, and merge the traffic participant information with an aggregation distance less than a preset distance into a traffic participant aggregation behavior;

[0036] A determination sub-module, configured to determine the traffic participant aggregation behavior and the risk confounding factor as driving risk source information.

[0037] Optionally, the first processing module includes:

[0038] A first prediction sub-module, configured to predict the traffic participant aggregation behavior by using multiple decision trees in a prediction model to obtain multiple prediction results of the traffic participant aggregation behavior;

[0039] A first mean sub-module, configured to perform mean processing on the multiple prediction results of the traffic participant aggregation behavior, and determine the prediction result after mean as the first prediction result of the traffic participant aggregation behavior affecting the driving risk result.

[0040] Optionally, the second processing module includes:

[0041] A second prediction sub-module, configured to predict the risk confounding factor by using multiple decision trees in an interference term model to obtain multiple prediction results of the risk confounding factor;

[0042] A second mean sub-module, configured to perform mean processing on the multiple prediction results of the risk confounding factor, and determine the prediction result after mean as the second prediction result of the risk confounding factor affecting the traffic participant aggregation behavior.

[0043] Optionally, the device further includes;

[0044] A data sampling module, configured to randomly sample the traffic participant aggregation behavior, the risk confounding factor, and the driving risk result to obtain a sampling sample set;

[0045] A model training module, configured to train a pre-constructed decision tree by using the sampling sample set to obtain a prediction model and an interference term model;

[0046] A model calling module, configured to call the prediction model and / or the interference term model according to the type of quantity to be predicted.

[0047] Optionally, the average causal effect module is configured such that the average causal effect of the traffic participant aggregation behavior on the driving risk result is calculated by the following formula:

[0048] ATE = Y c - g _pred - f _pred

[0049]

[0050]

[0051] where ATE is the average causal effect of the traffic participant aggregation behavior on the driving risk result, Y c is the driving risk result, g _pred is the first prediction result of the traffic participant aggregation behavior on the driving risk result, f _pred is the second prediction result of the risk confounding factor on the traffic participant aggregation behavior, T n (x) is the traffic participant aggregation behavior, W n (x) is the risk confounding factor, and N is the number of decision trees.

[0052] According to another aspect of the present invention, there is also provided an electronic device, including:

[0053] a processor;

[0054] a memory for storing instructions executable by the processor;

[0055] wherein the processor is configured to execute the instructions to implement the driving risk assessment method as described above.

[0056] According to another aspect of the present invention, there is also provided a readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the driving risk assessment method as described above are implemented.

[0057] The driving risk assessment method provided by the embodiments of the present invention obtains the driving risk source information and driving risk results of a vehicle, inputs the traffic participant aggregation behavior into a pre-trained prediction model to output a first prediction result of the influence of the traffic participant aggregation behavior on the driving risk results, inputs the risk confounding factor into a pre-trained interference item model to output a second prediction result of the influence of the risk confounding factor on the traffic participant aggregation behavior, calculates the residual between the driving risk results and the sum of the first prediction result and the second prediction result, and obtains the average causal effect of the traffic participant aggregation behavior on the driving risk results. By first eliminating the risk confounding factors in the driving risk sources and then calculating the average causal effect of the traffic participant aggregation behavior on the driving risk results, the embodiments of the present invention obtain the analysis result of the influence of the traffic participant aggregation behavior without the interference of confounding factors on the driving risk results, realize the accurate analysis of the causal relationship between traffic participants and driving risks, improve the accuracy of driving risk cause analysis, provide a reliable basis for the safe operation of the autonomous driving system, and further improve the stability and reliability of the autonomous driving system in complex environments.

[0058] The above description is only an overview of the technical solution of the present invention. In order to be able to understand the technical means of the present invention more clearly, it can be implemented according to the content of the description. And in order to make the above and other purposes, features and advantages of the present invention more obvious and understandable, the following specifically describes the embodiments of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] By reading the following detailed description of the preferred embodiments, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention. And throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:

[0060] Figure 1 is a flowchart of the steps of a driving risk assessment method provided by the embodiments of the present invention;

[0061] Figure 2 is Figure 1 a flowchart of step 101 in the driving risk assessment method provided by the embodiments of the present invention;

[0062] Figure 3 is Figure 1 a flowchart of step 102 in the driving risk assessment method provided by the embodiments of the present invention;

[0063] Figure 4 is Figure 1 a flowchart of step 103 in the driving risk assessment method provided by the embodiments of the present invention;

[0064] Figure 5It is a schematic structural diagram of a driving risk assessment device provided by an embodiment of the present invention;

[0065] Figure 6 It is a schematic structural diagram of an electronic device provided by an embodiment of the present invention. Specific embodiments

[0066] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the following will elaborate on the various embodiments of the present invention in conjunction with the accompanying drawings. However, those of ordinary skill in the art can understand that in various embodiments of the present invention, many technical details are provided to help readers better understand the present application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical solutions claimed in the present application can still be implemented. The division of the following embodiments is for convenience of description and should not constitute any limitation on the specific implementation of the present invention. The various embodiments can be combined and cross-referenced with each other on the premise of not being contradictory.

[0067] Referring to Figure 1 , a flowchart of the steps of a driving risk assessment method provided by an embodiment of the present invention is shown, and the method may include:

[0068] Step 101, obtain driving risk source information and driving risk results of a vehicle, where the driving risk source information includes traffic participant aggregation behavior and risk confounding factors.

[0069] In the embodiments of the present invention, the risk sources of driving risks in autonomous driving can be divided into explicit risks and implicit risks. Explicit risks include perceivable factors such as bad weather, traffic flow, and road conditions. Implicit risks involve factors such as interference and interaction among traffic participants. A change in the behavior of one participant may significantly affect the behavior of other participants. At the same time, explicit risks are not isolated and often interact with implicit risks, jointly affecting the safety of autonomous vehicles. Therefore, when analyzing the risk causes of autonomous vehicle operation, it is necessary to comprehensively consider the interaction and influence between different risks. Traditional driving risk cause analysis models do not consider the interference of risk confounding factors and directly conduct autonomous driving risk cause analysis on the behavior of traffic participants, which will cause a large error in driving risk cause analysis and further affect the accuracy of the model in analyzing the causal relationship between traffic participants and driving risks. Therefore, to solve the problem that risk confounding factors affect the accuracy of the driving risk cause analysis model results and the causal relationship between traffic participants and driving risks cannot be obtained, this embodiment adopts a two-stage machine learning method to first eliminate risk confounding factors and then calculate the causal effect of traffic participants on the operation risk of autonomous driving, so as to accurately analyze the causal relationship between traffic participants and driving risks and improve the accuracy of driving risk cause analysis.

[0070] It should be noted that in this embodiment, the risk source is a combination of the probability of an accident occurring and the severity of the resulting injury. The real road traffic system covers multiple core elements such as "humans-vehicles-roads-environments", forming a complex interaction network. During the driving process of autonomous vehicles, the various risk sources faced can be classified into four major categories: traffic participants, road conditions, traffic flow, and weather conditions. The purpose of this embodiment is to accurately analyze the causal factors of the impact of traffic participants on driving risk results. Among them, traffic participants mainly include pedestrians, vehicles, cyclists, traffic facilities, obstacles, animals, and miscellaneous items, etc. Road conditions mainly include road quality, road construction, etc. Traffic flow mainly includes traffic density, traffic flow, traffic patterns, etc. Traffic flow is affected by the spatio-temporal characteristics of traffic participants. Among them, severe weather in weather conditions is a typical risk confounding factor, and severe weather affects the behavior of traffic participants, the performance of vehicles, and the accuracy of environmental perception systems at the same time.

[0071] Specifically, first obtain the driving risk source information and driving risk results of the vehicle. Among them, the driving risk source information is divided into traffic participant aggregation behavior and risk confounding factors. Traffic participant aggregation behavior is the aggregation behavior of traffic participants such as pedestrians, vehicles, cyclists, traffic facilities, obstacles, animals, and miscellaneous items. The aggregation behavior is determined by the aggregation distance of traffic participants. The aggregation distance refers to the straight-line distance between traffic participants. If the aggregation distance exceeds the preset distance threshold, it is not regarded as traffic participant aggregation. If the aggregation distance is less than the preset distance, it is determined that traffic participants generate aggregation. The risk confounding factor is a factor that affects traffic participant aggregation behavior, mainly including weather conditions. Specifically, the driving risk source information and driving risk results can be obtained by collecting the operation data during the operation of autonomous vehicles. Among them, the operation data includes sensor data, operation logs, accident records, etc. Identify the driving risks that occur during the operation of autonomous vehicles and the risk sources that cause the risks from the collected operation data to obtain the driving risk source information and driving risk results of the vehicle.

[0072] It should be noted that the driving risk causal analysis model in this embodiment is a framework model for identifying, analyzing, and managing potential risk sources in a system or process. It is assumed that the driving risk source includes risk confounding factors and traffic participant aggregation behavior other than risk confounding factors. A two-stage machine learning method is adopted. First, eliminate the risk confounding factors in the driving risk source, and then calculate the causal effect of traffic participant aggregation behavior on driving risk results. By evaluating the impact of risk confounding factors on traffic participant aggregation behavior and then eliminating the impact of risk confounding factors on traffic participant aggregation behavior, the causal effect of unperturbed traffic participant aggregation behavior on the operation risk of autonomous vehicles is obtained, realizing the accurate analysis of the causal relationship between traffic participant aggregation behavior and driving risk results, and providing a reliable basis for the safe operation of autonomous driving systems.

[0073] Step 102: Input the traffic participant aggregation behavior into a pre-trained prediction model, and output a first prediction result of the influence of the traffic participant aggregation behavior on the driving risk result.

[0074] In the embodiment of the present invention, after obtaining the driving risk result, first evaluate the influence of the traffic participant aggregation behavior on the driving risk result, and then evaluate the risk confounding factors on the traffic participant aggregation behavior, so as to eliminate the risk confounding factors in the driving risk based on the driving risk result, and then calculate the causal effect of the traffic participant aggregation behavior on the driving risk result, so as to accurately analyze the causal relationship between the traffic participant aggregation behavior and the driving risk result. Specifically, input the traffic participant aggregation behavior into a pre-trained prediction model, and output a first prediction result of the influence of the traffic participant aggregation behavior on the driving risk result. Among them, the prediction model is a model used to predict the influence of a treatment variable on an outcome variable. The prediction model uses the random forest algorithm to establish the relationship between the treatment variable and the outcome variable. The treatment variable is used as the input, and the outcome variable is used as the target output. In this embodiment, the prediction model is used to predict the influence of the traffic participant aggregation behavior (treatment variable) on the driving risk result (outcome variable).

[0075] It should be noted that in this embodiment, the prediction model is pre-trained using the random forest algorithm. Specifically, relevant variables of traffic participants and driving risk results are collected from data sources such as traffic monitoring systems and sensor networks, the collected data is randomly sampled to generate a sampling sample set, decision trees are trained using the sampling sample set, and all decision trees are combined to obtain the prediction model, which will not be elaborated here one by one.

[0076] Specifically, input the traffic participant aggregation behavior into a pre-trained prediction model. Through multiple decision trees constructed in the prediction model, traverse and calculate each input traffic participant aggregation behavior variable, record the variable values corresponding to each node, process the variable values to obtain the prediction results of each decision tree, and calculate the average value of the prediction results obtained by the traffic participant aggregation behavior passing through the decision trees, which is the first prediction result of the influence of the traffic participant aggregation behavior on the driving risk result.

[0077] Step 103: Input the risk confounding factors into a pre-trained interference model, and output a second prediction result of the influence of the risk confounding factors on the traffic participant aggregation behavior.

[0078] In the embodiment of the present invention, after obtaining the first prediction result of the impact of the aggregation behavior of traffic participants on the driving risk result, since there are still risk confounding factors in the driving risk result that affect the aggregation behavior of traffic participants, therefore, by evaluating the prediction result of the impact of risk confounding factors on the aggregation behavior of traffic participants, the risk confounding factors in the driving risk are eliminated, and based on the driving risk result, the impact of the aggregation behavior of traffic participants on the driving risk result, and the impact of risk confounding factors on the aggregation behavior of traffic participants, the causal effect of the aggregation behavior of traffic participants on the driving risk result is calculated, so as to accurately analyze the causal relationship between the aggregation behavior of traffic participants and the driving risk result.

[0079] It should be noted that the pre-trained interference term model and the prediction model belong to the same type of model, which is a model used to predict the impact of a treatment variable on an outcome variable. The interference term model also uses the random forest algorithm to establish the relationship between the treatment variable and the outcome variable. The treatment variable is used as the input, and the outcome variable is used as the target output. The difference is that the interference term model is used to predict the impact of risk confounding factors (treatment variables) on the aggregation behavior of traffic participants (outcome variables). In the model training stage, relevant variables of traffic participants and risk confounding factors are collected from data sources such as traffic monitoring systems and sensor networks, the collected data is randomly sampled to generate a sampling sample set, and the sampling sample set is used to train decision trees, and all decision trees are combined to obtain the interference term model.

[0080] Specifically, the risk confounding factors are input into the pre-trained interference term model. Through the multiple decision trees constructed in the interference term model, each input quantity is traversed and calculated, the variable values corresponding to each node are recorded, the variable values are processed to obtain the prediction results of each decision tree, and the average value of the prediction results obtained by the risk confounding factors passing through the decision trees is calculated, and the second prediction result of the impact of the risk confounding factors on the aggregation behavior of traffic participants is output.

[0081] Step 104, calculate the residual between the driving risk result and the sum of the first prediction result and the second prediction result, and obtain the average causal effect of the aggregation behavior of traffic participants on the driving risk result.

[0082] In the embodiments of the present invention, based on the driving risk result, the impact of the traffic participant aggregation behavior on the driving risk result, and the impact of the risk confounding factor on the traffic participant aggregation behavior, the causal effect of the traffic participant aggregation behavior on the driving risk result is calculated. Specifically, the residual calculation is performed on the driving risk result, the output result of the prediction model, and the output result of the interference term model. The residual between the driving risk result and the sum of the first prediction result and the second prediction result is calculated. The calculated residual is the average causal effect of the traffic participant aggregation behavior on the driving risk result. The average treatment effect (ATE) measures the causal relationship between a certain risk factor (treatment variable) and the operation risk of the autonomous vehicle (result variable), and is used for causal analysis to identify and quantify the impact of risk factors on the safety of the autonomous driving system.

[0083] The driving risk assessment method provided by the embodiments of the present invention obtains the driving risk source information and the driving risk result of the vehicle, inputs the traffic participant aggregation behavior into a pre-trained prediction model to output the first prediction result of the impact of the traffic participant aggregation behavior on the driving risk result, inputs the risk confounding factor into a pre-trained interference term model to output the second prediction result of the impact of the risk confounding factor on the traffic participant aggregation behavior, and calculates the residual between the driving risk result and the sum of the first prediction result and the second prediction result to obtain the average causal effect of the traffic participant aggregation behavior on the driving risk result. By first eliminating the risk confounding factor in the driving risk source and then calculating the average causal effect of the traffic participant aggregation behavior on the driving risk result, the embodiments of the present invention obtain the analysis result of the impact of the traffic participant aggregation behavior on the driving risk result without the interference of the confounding factor, realize the accurate analysis of the causal relationship between the traffic participant and the driving risk, improve the accuracy of the driving risk causal analysis, provide a reliable basis for the safe operation of the autonomous driving system, and further improve the stability and reliability of the autonomous driving system in a complex environment.

[0084] Further, referring to Figure 2 shows Figure 1 the flowchart of step 101 in a driving risk assessment method provided, which is basically the same as the driving risk assessment method provided by the first embodiment of the present invention. Step 101 may include:

[0085] Step 1011, when it is determined that there is a driving risk during vehicle operation, obtain the driving risk result, as well as the traffic participant information and the risk confounding factor corresponding to the driving risk result.

[0086] Step 1012, calculate the aggregation distance in the traffic participant information, and merge the traffic participant information with an aggregation distance less than the preset distance into the traffic participant aggregation behavior.

[0087] Step 1013: Determine the traffic participant aggregation behavior and risk confounding factors as driving risk source information.

[0088] It should be noted that in the embodiments of the present invention, after confirming the existence of driving risks during vehicle operation, driving risk results, as well as traffic participant information and risk confounding factors corresponding to the driving risk results, are obtained from data sources such as traffic monitoring systems and sensor networks. Specifically, driving risks that occur during the operation of autonomous vehicles and the risk sources that cause the risks are identified from the collected operation data. After identifying and confirming the existence of driving risks during vehicle operation, driving risk results, as well as traffic participant information and risk confounding factors corresponding to the driving risk results, are obtained. Among them, the driving risk results can be specific risk types or levels. The traffic participant information includes pedestrians, vehicles, cyclists, traffic facilities, obstacles, animals, and miscellaneous items, etc. The risk confounding factors are factors that affect traffic participant aggregation behavior, mainly including weather conditions.

[0089] Specifically, due to the mutual influence among traffic participants, in this embodiment, the aggregation distance in the traffic participant information is calculated, and the traffic participant information with an aggregation distance less than the preset distance is merged into the traffic participant aggregation behavior. Among them, the aggregation behavior is determined by the aggregation distance of the traffic participants. The aggregation distance refers to the straight-line distance between traffic participants. If the aggregation distance exceeds the preset distance threshold, it is not considered as traffic participant aggregation. If the aggregation distance is less than the preset distance, it is determined that the traffic participants generate aggregation. Specifically, the spatial relationship between traffic participants can be analyzed through data collected by sensors, cameras, etc., the distance between traffic participants can be measured or calculated, and the participant information with a distance less than the preset distance value is merged and regarded as a group behavior. In this embodiment, the value of the preset distance in the aggregation behavior determination is not specifically limited.

[0090] In the embodiments of the present invention, by analyzing the behaviors and risk factors of traffic participants in driving risks, the sources of driving risks are accurately identified, and the traffic participant aggregation behavior and risk confounding factors to be evaluated are obtained, providing a basis for subsequent risk cause analysis and decision-making.

[0091] Further, referring to Figure 3 , shows Figure 1 The flowchart of step 102 in a provided driving risk assessment method, which is basically the same as the driving risk assessment method provided in the first embodiment of the present invention. Step 102 may include:

[0092] Step 1021: Use multiple decision trees in the prediction model to predict the traffic participant aggregation behavior and obtain multiple prediction results of the traffic participant aggregation behavior.

[0093] Step 1022, perform mean processing on multiple prediction results of the traffic participant aggregation behavior, and determine the prediction result after mean as the first prediction result of the influence of the traffic participant aggregation behavior on the driving risk result.

[0094] It should be noted that in the embodiment of the present invention, multiple decision trees in the prediction model are used to predict the traffic participant aggregation behavior, and prediction results corresponding to the multiple decision trees are obtained. Based on the number of decision trees, mean processing is performed on multiple prediction results of the influence of the traffic participant aggregation behavior on the driving risk result, and the prediction result after mean is determined as the first prediction result of the influence of the traffic participant aggregation behavior on the driving risk result. For each decision tree, the traffic participant aggregation behavior is input to obtain a prediction result. The prediction results of all decision trees are collected, and the mean of the prediction results of all decision trees is calculated, and the mean result is determined as the first prediction result.

[0095] Specifically, the feature matrix X is included in the prediction model c , and random sampling is performed on the feature matrix X c to obtain a sampling sample set D. where x i is the feature vector of the sample, y i is the label of the sample. According to the sampling sample set, the traffic participant aggregation behavior T c and the risk confounding factor W c , a decision tree is constructed. N decision trees N = {N1, N2,..., N N}, and the t-th decision tree is N t , N t = buildTree(D, T t , W c ). For a new sample x, it is input into the constructed decision tree for prediction to obtain the prediction values of each decision tree, and the final prediction result is obtained by averaging, that is, the average of the prediction results of each decision tree is used as the final prediction result:

[0096]

[0097] where g _pred is the first prediction result of the influence of the traffic participant aggregation behavior on the driving risk result, T n (x) is the traffic participant aggregation behavior, and N is the number of decision trees.

[0098] In the embodiment of the present invention, by using multiple decision trees to predict the traffic participant aggregation behavior and calculating the mean, the influence of the traffic participant aggregation behavior on the driving risk result can be captured more accurately, and the accuracy of causal analysis can be improved.

[0099] Further, referring to Figure 4 , it showsFigure 1 The flowchart of step 103 in a provided driving risk assessment method, which is basically the same as the driving risk assessment method provided in the first embodiment of the present invention. Step 103 may include:

[0100] Step 1031, using multiple decision trees in the interference item model to predict the risk confounding factors, and obtaining multiple prediction results of the risk confounding factors.

[0101] Step 1032, performing a mean processing on the multiple prediction results of the risk confounding factors, and determining the prediction result after mean as the second prediction result of the influence of the risk confounding factors on the aggregation behavior of traffic participants.

[0102] In this embodiment, multiple decision trees in the interference item model are used to predict the risk confounding factors, and corresponding prediction results of the multiple decision trees are obtained. Specifically, for each decision tree, the risk confounding factors are input, and the prediction result of the influence of the risk confounding factors on the aggregation behavior of traffic participants is obtained. The prediction results of all decision trees are collected, and based on the number of decision trees, a mean processing is performed on the prediction results of the multiple decision trees, and the prediction result after mean is determined as the second prediction result of the influence of the risk confounding factors on the aggregation behavior of traffic participants, that is, calculating the mean of the prediction results of all decision trees, and determining the mean result as the second prediction result.

[0103] It should be noted that for the risk confounding factor W c , it is input into the constructed decision tree for prediction to obtain the predicted values of each decision tree, and the final prediction result is obtained by averaging, that is, the average of the prediction results of each decision tree is used as the final prediction processing result:

[0104]

[0105] where f _pred is the second prediction result of the influence of the risk confounding factors on the aggregation behavior of traffic participants, W n (x) is the risk confounding factor, and N is the number of decision trees.

[0106] In the embodiment of the present invention, by using multiple decision trees to predict the risk confounding factors and calculating the mean, the influence of the risk confounding factors on the aggregation behavior of traffic participants is accurately determined, improving the accuracy of causal analysis.

[0107] Specifically, after obtaining the driving risk source information and driving risk results of the vehicle, where the driving risk source information includes the aggregation behavior of traffic participants and risk confounding factors, it further includes;

[0108] First, randomly sample the aggregation behavior of traffic participants, risk confounding factors, and driving risk results to obtain a sampling sample set;

[0109] Secondly, use the sampled sample set to train a pre-constructed decision tree to obtain a prediction model and a confounding model.

[0110] Secondly, according to the type of variable to be predicted, call the prediction model and / or the confounding model.

[0111] It should be noted that in the above steps of this embodiment, relevant variables, risk confounding factors, and corresponding driving risk results of traffic participants are collected from data sources such as traffic monitoring systems and sensor networks, the collected data is randomly sampled to generate a sampled sample set, the sampled sample set is used to train a decision tree, and the trained decision trees are combined into a random forest to obtain a prediction model and a confounding model. Among them, the prediction model is used to predict the impact of the aggregation behavior (treatment variable) of traffic participants on the driving risk result (outcome variable), and the confounding model is used to predict the impact of risk confounding factors (treatment variable) on the aggregation behavior (outcome variable) of traffic participants. The specific steps are as follows: for each sampled sample set, train a decision tree, combine all decision tree models into a random forest, obtain a prediction model and a confounding model according to the prediction requirements, and call the corresponding model for risk prediction according to the type of variable to be predicted. For example, if the variable to be predicted is the aggregation behavior of traffic participants, then call the prediction model; if the variable to be predicted is a risk confounding factor, then call the confounding model.

[0112] The prediction model and the confounding model obtained through training in the embodiment of the present invention can respectively predict the complex relationships among traffic participants, risk confounding factors, and driving risk results, improve the accuracy of risk prediction, and achieve accurate evaluation of the aggregation behavior of traffic participants and risk confounding factors.

[0113] Specifically, step 104 calculates the residual of the driving risk result and the sum of the first prediction result and the second prediction result to obtain the average causal effect of the aggregation behavior of traffic participants on the driving risk result. The average causal effect of the aggregation behavior of traffic participants on the driving risk result is calculated by the following formula:

[0114] ATE = Y c - g _pred - f _pred

[0115]

[0116] where ATE is the average causal effect of the aggregation behavior of traffic participants on the driving risk result, Y c is the driving risk result, g _pred is the first prediction result of the aggregation behavior of traffic participants on the driving risk result, f _pred is the second prediction result of risk confounding factors on the aggregation behavior of traffic participants, Tn (x) is the aggregation behavior of traffic participants, W n (x) is the risk confounding factor, and N is the number of decision trees.

[0117] Specifically, the residual of the driving risk result with respect to the sum of the prediction model result g_pred and the interference term model result f_pred obtained in the fitting stage is calculated. The residual is the driving risk result Y c which is the difference between g_pred and f_pred. By eliminating the risk confounding factor in the driving risk source, and then calculating the average causal effect of the aggregation behavior of traffic participants on the driving risk result, the analysis result of the aggregation behavior of traffic participants without the interference of the confounding factor on the driving risk result is obtained, so as to accurately analyze the causal relationship between traffic participants and driving risk.

[0118] Refer to Figure 5 , which shows a schematic structural diagram of a driving risk assessment device 200 provided by an embodiment of the present invention. The device includes:

[0119] An acquisition module 201, configured to acquire driving risk source information and driving risk results of a vehicle. Among them, the driving risk source information includes the aggregation behavior of traffic participants and risk confounding factors;

[0120] A first processing module 202, configured to input the aggregation behavior of traffic participants into a pre-trained prediction model, and output a first prediction result of the influence of the aggregation behavior of traffic participants on the driving risk result;

[0121] A second processing module 203, configured to input the risk confounding factor into a pre-trained interference term model, and output a second prediction result of the influence of the risk confounding factor on the aggregation behavior of traffic participants;

[0122] An average causal effect module 204, configured to calculate the residual of the driving risk result with respect to the sum of the first prediction result and the second prediction result, and obtain the average causal effect of the aggregation behavior of traffic participants on the driving risk result.

[0123] Further, the acquisition module 201 includes:

[0124] An acquisition sub-module, configured to acquire the driving risk result, as well as the traffic participant information and risk confounding factors corresponding to the driving risk result, when it is determined that there is a driving risk during vehicle operation;

[0125] An aggregation sub-module, configured to calculate the aggregation distance in the traffic participant information, and merge the traffic participant information with an aggregation distance less than a preset distance into the aggregation behavior of traffic participants;

[0126] A determination sub-module, configured to determine the traffic participant aggregation behavior and the risk confounding factor as driving risk source information.

[0127] Further, the first processing module 202 includes:

[0128] A first prediction sub-module, configured to use multiple decision trees in a prediction model to predict the traffic participant aggregation behavior, and obtain multiple prediction results of the traffic participant aggregation behavior;

[0129] A first mean sub-module, configured to perform mean processing on the multiple prediction results of the traffic participant aggregation behavior, and determine the prediction result after mean as the first prediction result of the traffic participant aggregation behavior affecting the driving risk result.

[0130] Further, the second processing module 203 includes:

[0131] A second prediction sub-module, configured to use multiple decision trees in an interference term model to predict the risk confounding factor, and obtain multiple prediction results of the risk confounding factor;

[0132] A second mean sub-module, configured to perform mean processing on the multiple prediction results of the risk confounding factor, and determine the prediction result after mean as the second prediction result of the risk confounding factor affecting the traffic participant aggregation behavior.

[0133] Optionally, the device further includes;

[0134] A data sampling module, configured to randomly sample the traffic participant aggregation behavior, the risk confounding factor, and the driving risk result, and obtain a sampling sample set;

[0135] A model training module, configured to use the sampling sample set to train a pre-constructed decision tree, and obtain a prediction model and an interference term model;

[0136] A model calling module, configured to call the prediction model and / or the interference term model according to the type of the quantity to be predicted.

[0137] Further, the average causal effect module 204 is configured to calculate the average causal effect of the traffic participant aggregation behavior on the driving risk result through the following formula:

[0138] ATE = Y c - g _pred - f _pred

[0139]

[0140] Among them, ATE is the average causal effect of the aggregation behavior of traffic participants on the driving risk result, and Y c is the driving risk result, and g _pred is the first prediction result of the aggregation behavior of traffic participants on the driving risk result, and f _pred is the second prediction result of the risk confounding factor on the aggregation behavior of traffic participants, and T n (x) is the aggregation behavior of traffic participants, and W n (x) is the risk confounding factor, and N is the number of decision trees.

[0141] The driving risk assessment device provided by the embodiment of the present invention obtains the driving risk source information and driving risk result of the vehicle, inputs the aggregation behavior of traffic participants into a pre-trained prediction model, outputs the first prediction result of the aggregation behavior of traffic participants on the driving risk result, inputs the risk confounding factor into a pre-trained interference term model, outputs the second prediction result of the risk confounding factor on the aggregation behavior of traffic participants, calculates the residual between the driving risk result and the sum of the first prediction result and the second prediction result, and obtains the average causal effect of the aggregation behavior of traffic participants on the driving risk result. By first eliminating the risk confounding factor in the driving risk source and then calculating the average causal effect of the aggregation behavior of traffic participants on the driving risk result, the embodiment of the present invention obtains the analysis result of the aggregation behavior of traffic participants on the driving risk result without the interference of the confounding factor, realizes the accurate analysis of the causal relationship between traffic participants and driving risk, improves the accuracy of the driving risk cause analysis, provides a reliable basis for the safe operation of the autonomous driving system, and further improves the stability and reliability of the autonomous driving system in a complex environment.

[0142] Referring to Figure 6 , the embodiment of the present invention also provides an electronic device, as Figure 6 shown, including a processor 301, a communication interface 302, a memory 303, and a communication bus 304. Among them, the processor 301, the communication interface 302, and the memory 303 communicate with each other through the communication bus 304,

[0143] the processor 301, and the memory 303 for storing processor-executable instructions;

[0144] Among them, the processor 301 is configured to execute the instructions to implement the driving risk assessment method as described above.

[0145] The communication bus mentioned in the above terminal may be a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience of representation, only a thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0146] The communication interface is used for communication between the above terminal and other devices.

[0147] The memory may include a Random Access Memory (RAM), or may also include a non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one storage device located far from the aforementioned processor.

[0148] The above-mentioned processor may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0149] In another embodiment provided by the present invention, a computer-readable storage medium is also provided. A computer program is stored on the readable storage medium, and when the computer program is executed by a processor, the driving risk assessment method described in any one of the above embodiments is implemented.

[0150] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present invention are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wire (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or a data center that integrates one or more available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)).

[0151] It should be noted that in this document, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including", or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article, or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article, or device. Without further limitation, an element defined by the statement "including an..." does not exclude the existence of additional identical elements in the process, method, article, or device including the element.

[0152] Each embodiment in this specification is described in a related manner. The same or similar parts among the embodiments can be referred to each other, and the differences between each embodiment and other embodiments are emphasized. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiment.

[0153] The above are only the preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention are all included within the protection scope of the present invention.

Claims

1. A driving risk assessment method, characterized in that: The method comprises: Acquiring driving risk source information and driving risk results of the vehicle, wherein the driving risk source information includes traffic participant aggregation behavior and risk confounding factors; Inputting the traffic participant gathering behavior into a pre-trained prediction model, and outputting a first prediction result of the traffic participant gathering behavior affecting the driving risk result; Inputting the risk confounding factor into a pre-trained interference term model, and outputting a second prediction result of the risk confounding factor affecting the gathering behavior of the traffic participants; The residual of the driving risk result and the sum of the first prediction result and the second prediction result is calculated to obtain the average causal effect of the traffic participants' gathering behavior on the driving risk result.

2. The method according to claim 1, characterized in that: The obtaining of the driving risk source information and driving risk results of the vehicle, wherein the driving risk source information includes the gathering behavior of traffic participants and risk confounding factors, includes: When it is determined that there is a driving risk in the operation of the vehicle, obtaining a driving risk result, and traffic participant information and risk confounding factors corresponding to the driving risk result; Calculating the gathering distance in the traffic participant information, and merging the traffic participant information whose gathering distance is less than a preset distance into traffic participant gathering behavior; The traffic participant aggregation behavior and the risk confounding factor are determined as driving risk source information.

3. The method according to claim 1, characterized in that The step of inputting the traffic participant gathering behavior into a pre-trained prediction model and outputting a first prediction result of the traffic participant gathering behavior affecting the driving risk result includes: Using multiple decision trees in the prediction model to predict the gathering behavior of the traffic participants, and obtaining multiple prediction results of the gathering behavior of the traffic participants; A plurality of prediction results of the gathering behavior of the traffic participants are averaged, and the averaged prediction result is determined as a first prediction result of the impact of the gathering behavior of the traffic participants on the driving risk result.

4. The method according to claim 1, characterized in that: The step of inputting the risk confounding factor into a pre-trained interference term model and outputting a second prediction result of the risk confounding factor affecting the gathering behavior of traffic participants includes: Using multiple decision trees in the interference term model to predict the risk confounding factor to obtain multiple prediction results of the risk confounding factor; A plurality of prediction results of the risk confounding factor are averaged, and the averaged prediction result is determined as a second prediction result of the risk confounding factor affecting the gathering behavior of the traffic participants.

5. The method according to claim 3 or 4, characterized in that: The obtaining of the driving risk source information and driving risk result of the vehicle, wherein the driving risk source information includes the gathering behavior of traffic participants and the risk confounding factor, and further includes: Randomly sampling the traffic participant aggregation behavior, the risk confounding factor, and the driving risk result to obtain a sampling sample set; The sample set is used to train a pre-constructed decision tree to obtain a prediction model and an interference term model; The prediction model and / or the interference term model is called according to the type of the quantity to be predicted.

6. The method according to claim 1, characterized in that The residual of the driving risk result and the sum of the first prediction result and the second prediction result is calculated to obtain the average causal effect of the traffic participant gathering behavior on the driving risk result. The average causal effect of the traffic participant gathering behavior on the driving risk result is calculated by the following formula: ATE=Y c -g _pred -f _pred Among them, ATE is the average causal effect of traffic participants’ clustering behavior on driving risk outcomes, Y c is the driving risk result, g _pred is the first prediction result of the impact of traffic participants’ gathering behavior on driving risk outcomes, f _pred is the second prediction result of the risk confounding factor affecting the gathering behavior of traffic participants, T n (x) is the gathering behavior of traffic participants, W n (x) is the risk confounder and N is the number of decision trees.

7. A driving risk assessment device, characterized in that: The device comprises: An acquisition module, used to acquire driving risk source information and driving risk results of the vehicle, wherein the driving risk source information includes traffic participant aggregation behavior and risk confounding factors; A first processing module, configured to input the traffic participant gathering behavior into a pre-trained prediction model, and output a first prediction result of the traffic participant gathering behavior affecting the driving risk result; A second processing module is used to input the risk confounding factor into a pre-trained interference term model, and output a second prediction result of the risk confounding factor affecting the gathering behavior of the traffic participants; The average causal effect module is used to calculate the residual of the driving risk result and the sum of the first prediction result and the second prediction result to obtain the average causal effect of the traffic participants' gathering behavior on the driving risk result.

8. The device according to claim 7, characterized in that The acquisition module comprises: An acquisition submodule, for acquiring a driving risk result, and traffic participant information and risk confounding factors corresponding to the driving risk result, when determining that there is a driving risk in the operation of the vehicle; A gathering submodule, used for calculating the gathering distance in the traffic participant information, and merging the traffic participant information whose gathering distance is less than a preset distance into the traffic participant gathering behavior; The determination submodule is used to determine the traffic participant aggregation behavior and the risk confounding factor as driving risk source information.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to execute the instructions to implement the driving risk assessment method as described in any one of claims 1 to 6.

10. A readable storage medium, characterized in that: The readable storage medium stores a computer program, and when the computer program is executed by a processor, the driving risk assessment method according to any one of claims 1 to 6 is implemented.

Images