An l3-level autonomous driving takeover process safety evaluation method based on iaHP-EWM-LDM

Abstract

The application provides an L3 level automatic driving takeover process safety evaluation method based on IAHP-EWM-LDM, which comprehensively considers the driver visual characteristics, the driver steering behavior and the vehicle running state in the whole takeover process, proposes a takeover process safety representation parameter, screens the parameter through a variation coefficient method and a Spearman correlation discrimination method to obtain an evaluation index, obtains an index subjective weight by using an improved analytic hierarchy process, obtains an index objective weight by using an entropy weight method, and obtains an index comprehensive weight by using a maximum level difference method to combine the subjective weight and the objective weight, so as to propose an index weight obtaining method combining the improved analytic hierarchy process, the entropy weight method and the maximum level difference method. Therefore, a takeover process safety evaluation model is constructed to evaluate the takeover process. Compared with the prior art, the application analyzes the whole process of the driver taking over the vehicle, and increases the index of risk perception. On the basis of analyzing the objective data to obtain information, the subjective expert experience is combined to realize the comprehensive evaluation of the takeover process safety.

Description

Technical Field

[0001] This invention relates to a safety evaluation method for L3 level autonomous driving takeover process based on IAHP-EWM-LDM, belonging to the fields of driving behavior and intelligent transportation. Background Technology

[0002] Existing studies on driver takeover safety assessments primarily focus on multiple factors as independent variables, analyzing the differences among various safety evaluation indicators. This approach fails to provide a comprehensive evaluation of driver takeover safety. Evaluation indicators often include reaction time, minimum time-to-control (TTC), and speed. These indicators only describe a driver's reaction and vehicle control abilities, failing to describe their risk perception capabilities, which are crucial for safety. Furthermore, relying solely on statistical analysis of indicator data as the evaluation standard allows for significant variations in results depending on fluctuations in objective data. In addition, current research emphasizes the driver's performance during the takeover request, but subsequent driving performance also significantly impacts driving safety. Summary of the Invention

[0003] To address the issues of incomplete indicator selection and significant fluctuations in evaluation results due to objective data in the safety assessment of takeover processes, this invention proposes a safety assessment method for takeover processes based on an improved analytic hierarchy process and entropy weight method.

[0004] This method comprehensively considers perception, decision-making, and manipulation during the takeover process. It extracts 13 evaluation parameters from three aspects: risk perception, risk avoidance manipulation, and takeover performance. These parameters are screened using the coefficient of variation method and Spearman's correlation discriminant method to obtain evaluation indicators. The subjective weights of the indicators are obtained using the improved analytic hierarchy process (IAHP), the objective weights using the entropy weight method, and the differential maximization method (LDM) is used to combine the subjective and objective weights to obtain the comprehensive weights of the indicators. Therefore, a method combining the improved analytic hierarchy process (IAHP), the entropy weight method (EWM), and the differential maximization method (LDM) is proposed to establish a safety evaluation indicator system for the takeover process and to evaluate the safety of the takeover process.

[0005] The principle behind using this method to evaluate the safety of the takeover process is explained below.

[0006] A complete takeover process involves the driver perceiving the level of road risk and making the correct decision based on that level. Therefore, the safety evaluation of the takeover process should include an analysis of the driver's perceived risk level and operational ability, as well as an assessment of the vehicle's control ability after takeover. The selection of indicators should consider information carrying capacity, and indicators with a coefficient of variation less than 15% should be eliminated using the coefficient of variation method. Furthermore, since the paper subsequently uses the analytic hierarchy process (AHP) to calculate expert weights, and AHP requires that evaluation indicators be independent of each other, correlation analysis of the parameters is necessary to simplify the judgment information of the indicators and improve the accuracy of expert judgment.

[0007] The Analytic Hierarchy Process (AHP) was first proposed by the renowned American operations researcher L.T. Saaty. Its core idea is to decompose the causes of a problem into different levels based on the relationships between them, then establish a judgment matrix according to the relative importance of factors at the same level, and determine the weight of each factor by calculating the eigenvalues ​​of the judgment matrix. Traditional AHP often uses the nine-scale method to illustrate the relative importance of factors. Without considering the nine-scale method, it is difficult to clearly classify importance, and expert subjective judgments often contain contradictions, causing the judgment matrix to fail consistency tests. Subsequent consistency tests require further subjective adjustments.

[0008] To address this issue, a five-scale method was chosen to improve the analytic hierarchy process, ensuring clear comparison of importance while maintaining the discriminative power between factors. However, subjective weights are heavily influenced by expert experience, potentially leading to unreasonable weighting of indicators. To overcome the shortcomings of solely subjective weighting, objective weights need to be added for adjustment. Furthermore, considering the advantages of both weighting methods and the need for more interpretable results, the difference maximization method was selected to combine subjective and objective weights.

[0009] In summary, based on the safety evaluation indicators and their comprehensive weights, a takeover safety evaluation indicator system can be established to evaluate the takeover process.

[0010] A safety evaluation method for L3 autonomous driving takeover process based on IAHP-EWM-LDM, characterized by the following steps:

[0011] Step 1: Propose security characterization parameters;

[0012] Thirteen safety characterization parameters were selected, including mean fixation time (Fix_time), mean sac time (Sac_time), maximum pupil area difference (Diff_pupil), pupil area change rate (Rate_pupil); takeover time (Mani_time), control point (TTCMani_TTC), brake depth standard deviation (Std_brake), throttle depth standard deviation (Std_gas), steering wheel angle standard deviation (Std_wheel); lateral offset standard deviation (Std_lp), lateral acceleration standard deviation (Std_a_x), vehicle speed standard deviation (Std_speed), and longitudinal acceleration standard deviation (Std_a_y).

[0013] Step 2: Screen safety characterization parameters and determine safety evaluation indicators;

[0014] Calculate the coefficient of variation for each parameter to assess its information-carrying capacity. A larger coefficient of variation indicates a stronger information-carrying capacity; indicators with a coefficient of variation less than 15% should be discarded. Use the Spearman correlation test to analyze the correlation between parameters. If the correlation coefficient is greater than 0.8, it indicates excessive information overlap between indicators, and one needs to be removed. Therefore, the indicators are determined based on the coefficient of variation method and the Spearman correlation test.

[0015] Step 3: Calculate the weights of the safety evaluation indicators;

[0016] The weights of the indicators given by each expert were calculated using the improved analytic hierarchy process (AHP). The entropy weights of the evaluation indicators were then calculated by combining the driver takeover safety evaluation indicator system with the collected indicator data. To effectively combine expert opinions and data information, and considering the advantages of both, the step difference maximization method was selected, and MATLAB software was used to combine subjective and objective weights to obtain the combined weights of each indicator.

[0017] The process of combining subjective and objective weights is as follows:

[0018] (1) Constructing the weight matrix

[0019]

[0020] Among them, a i1 To improve the weighting of the i-th evaluation index in the analytic hierarchy process, a i2 This refers to the weighting of the i-th evaluation index using the entropy weighting method.

[0021] (2) Determine the range of combined weights

[0022] The combined weights a = (a1, a2, ..., a2) can be determined from the weight matrix A. k The range of the interval: in:

[0023]

[0024]

[0025] (3) Determine the optimal combination weights

[0026] The optimal combination of weights is determined with the goal of maximizing the variance of scores among n evaluation objects, so as to achieve a high degree of score differentiation among the evaluation objects.

[0027] ① Calculate the mean value of n evaluation objects under the i-th evaluation indicator:

[0028]

[0029] ② Calculate the variance of the comprehensive evaluation results of n evaluation objects under the i-th evaluation index:

[0030]

[0031] ③ Calculate the maximum value of the sum of variances of the comprehensive evaluation results under all evaluation indicators:

[0032]

[0033] And it satisfies the following constraints:

[0034]

[0035] In the formula: n is the number of evaluation objects, i.e., the number of takeover processes; s 2 The sum of variances of the comprehensive evaluation results under all evaluation indicators; a i The combined weights for evaluation index i; x j The value of the evaluation object.

[0036] Step 4: Conduct a safety assessment of each takeover process.

[0037] This invention proposes an evaluation method for autonomous driving takeover processes and uses this method to evaluate 655 takeover processes obtained from driving simulation experiments. Compared with existing technologies, this invention has the following advantages:

[0038] (1) This method takes into account the driver’s visual characteristics, driver’s operating behavior and vehicle operating status throughout the entire takeover process, ensuring the comprehensiveness of the indicator selection.

[0039] (2) When setting weights, while ensuring the representation of objective data information, subjective expert experience is integrated, and the subjective and objective weights are combined using the more interpretable differential maximization method, so that the method can comprehensively, reasonably and scientifically evaluate the safety of the takeover process. Attached Figure Description

[0040] Figure 1 For characterizing the parameter graph;

[0041] Figure 2 Flowcharts were created to define the evaluation metrics;

[0042] Figure 3 Flowchart for setting the weights of evaluation indicators;

[0043] Figure 4 This is a flowchart of the safety evaluation method for the takeover process involved in this invention. Detailed Implementation

[0044] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0045] Taking into account risk perception, risk avoidance maneuvering, and takeover performance, 13 characterization parameters are extracted from the safety evaluation perspective. These include: mean fixation time (Fix_time), mean sac time (Sac_time), maximum pupil area difference (Diff_pupil), pupil area change rate (Rate_pupil); takeover maneuvering time (Mani_time), maneuvering point time (TTC) (Mani_TTC), standard deviation of braking depth (Std_brake), standard deviation of throttle depth (Std_gas), standard deviation of steering wheel angle (Std_wheel); standard deviation of lateral offset (Std_lp), standard deviation of lateral acceleration (Std_a_x), standard deviation of vehicle speed (Std_speed), and standard deviation of longitudinal acceleration (Std_a_y). The system architecture is as follows: Figure 1 As shown. The parameters were screened using the coefficient of variation method and Spearman correlation. The screening process is as follows: Figure 2 As shown, since all parameters passed the screening, 13 evaluation indicators were finally determined.

[0046] An expert questionnaire was designed based on the improved analytic hierarchy process (AHP) five-scale value standard. The questionnaire results were used to calculate the weights of the indicators given by each expert. Combined with the driver takeover safety evaluation indicator system and the collected indicator data, the objective weights of the indicators were calculated. The step difference maximization method was selected, and MATLAB software was used to combine subjective and objective weights to obtain the combined weights of each indicator. For example... Figure 3 As shown, the weights of each security evaluation index are obtained.

[0047] The scores for each takeover process can be calculated based on the safety evaluation indicators and their corresponding weights, thereby achieving a safety evaluation of the takeover process.

[0048] The safety evaluation method for the takeover process described in this invention is as follows: Figure 4 As shown, the specific steps include the following:

[0049] Step 1: Propose security characterization parameters;

[0050] Step 2: Screening characterization parameters and determining safety evaluation indicators;

[0051] Step 3: Calculate the weights of the safety evaluation indicators;

[0052] Step 4: Conduct a safety assessment of each takeover process;

[0053] The method for determining the safety evaluation indicators is as follows: The safety characterization parameters are tested using the coefficient of variation method, and those with a coefficient of variation less than 15% are removed. The Spearman correlation test is also performed on the safety characterization parameters; if the correlation coefficient is greater than 0.8, it indicates that the information overlap between the indicators is too high, and one needs to be removed. All 13 safety characterization parameters passed the test and were therefore retained, thus being determined as the safety evaluation indicators.

[0054] The method for determining the weights of safety evaluation indicators is as follows: an expert questionnaire is designed based on the five-scale value standard of the improved analytic hierarchy process. The weights of the indicators given by each expert are calculated based on the questionnaire results. The objective weights of the indicators are calculated by combining the driver takeover safety evaluation indicator system and the collected indicator data. The step difference maximization method is selected and MATLAB software is used to combine the subjective and objective weights to obtain the combined weights of each indicator.

[0055] The process of combining subjective and objective weights is as follows:

[0056] (1) Constructing the weight matrix

[0057]

[0058] Among them, a i1 Improving the weighting of the i-th evaluation index using the analytic hierarchy process, a i2 The entropy weight method assigns weights to the i-th evaluation index.

[0059] (2) Determine the range of combined weights

[0060] The combined weights a = (a1, a2, ..., a2) can be determined from the weight matrix A. k The range of the interval: in:

[0061]

[0062]

[0063] (3) Determine the optimal combination weights

[0064] The optimal combination of weights is determined with the goal of maximizing the variance of scores among n evaluation objects, so as to achieve a high degree of score differentiation among the evaluation objects.

[0065] ① Calculate the mean value of n evaluation objects under the i-th evaluation indicator:

[0066]

[0067] ② Calculate the variance of the comprehensive evaluation results of n evaluation objects under the i-th evaluation index:

[0068]

[0069] ③ Calculate the maximum value of the sum of variances of the comprehensive evaluation results under all evaluation indicators:

[0070]

[0071] And it satisfies the following constraints:

[0072]

[0073] In the formula: n is the number of evaluation objects, i.e., the number of takeover processes; s 2 The sum of variances of the comprehensive evaluation results under all evaluation indicators; a i The combined weights for evaluation index i; x j The value of the evaluation object.

[0074] This invention comprehensively considers the driver's visual characteristics, driving behavior, and vehicle operating status throughout the entire takeover process. It extracts evaluation parameters from risk perception, hazard avoidance maneuvers, and takeover performance, resulting in comprehensive, reasonable, and scientific evaluation parameters. The weighting effectively combines expert experience and data information, making the evaluation results of takeover safety more objective and accurate.

Claims

1. A safety evaluation method for L3 level autonomous driving takeover process based on IAHP-EWM-LDM, characterized by the following steps: Step 1: Propose security characterization parameters; Thirteen safety characterization parameters were selected, including mean fixation time (Fix_time), mean sac time (Sac_time), maximum pupil area difference (Diff_pupil), pupil area change rate (Rate_pupil); takeover time (Mani_time), control point (TTCMani_TTC), brake depth standard deviation (Std_brake), throttle depth standard deviation (Std_gas), steering wheel angle standard deviation (Std_wheel); lateral offset standard deviation (Std_lp), lateral acceleration standard deviation (Std_a_x), vehicle speed standard deviation (Std_speed), and longitudinal acceleration standard deviation (Std_a_y). Step 2: Screen safety characterization parameters and determine safety evaluation indicators; Calculate the coefficient of variation for each parameter to assess its information-carrying capacity. The larger the coefficient of variation, the stronger the information-carrying capacity of the indicator. Indicators with a coefficient of variation of less than 15% should be eliminated. Use the Spearman correlation test to analyze the correlation between parameters. If the correlation coefficient is greater than 0.8, it indicates that the information overlap between indicators is too high, and one needs to be eliminated. Therefore, the indicators were determined based on the coefficient of variation method and Spearman correlation discrimination method. Step 3: Calculate the weights of the safety evaluation indicators; The weights of the indicators given by each expert were calculated based on the improved analytic hierarchy process. The entropy weights of the evaluation indicators were calculated by combining the driver takeover safety evaluation indicator system and the collected indicator data. The step difference maximization method was selected and MATLAB software was used to combine subjective and objective weights to obtain the combined weights of each indicator. The process of combining subjective and objective weights is as follows: (1) Constructing the weight matrix Among them, a i1 Improving the weighting of the i-th evaluation index using the analytic hierarchy process, a i2 The entropy weight method is used to assign weights to the i-th evaluation index; (2) Determine the range of combined weights The combined weights a = (a1, a2, ..., a2) can be determined from the weight matrix A. k The range of the interval: in: (3) Determine the optimal combination weights With the goal of maximizing the variance of scores among n evaluation objects, the optimal combination of weights is determined to ensure high score differentiation among the evaluation objects. ① Calculate the mean value of n evaluation objects under the i-th evaluation indicator: ② Calculate the variance of the comprehensive evaluation results of n evaluation objects under the i-th evaluation index: ③ Calculate the maximum value of the sum of variances of the comprehensive evaluation results under all evaluation indicators: And it satisfies the following constraints: In the formula: n is the number of evaluation objects, i.e., the number of takeover processes; s 2 The sum of variances of the comprehensive evaluation results under all evaluation indicators; a i The overall weight of evaluation index i; x j The value of the evaluation object; Step 4: Conduct a safety assessment of each takeover process; Based on the normalized raw values ​​and the comprehensive weights of the indicators obtained by the step difference maximization method, the safety evaluation score Z of the j-th takeover process is calculated. j : In the formula: x ij Let be the evaluation score of the j-th takeover process under the i-th evaluation index.

Images