Scene parameter-based automatic driving system safety analysis method and system

Abstract

The invention provides an automatic driving system safety analysis method and system based on scene parameters, and relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining scene parameter data; based on the scene parameter data, obtaining a parameter vector instance and a characteristic quantity of the automatic driving system under the parameter vector instance; training the linear model to obtain a target model; the target model determines the parameter vector instances and the feature quantities which are ranked in the front according to constraint conditions, and target parameters are obtained; any two parameters in the target parameters are combined to obtain a target parameter combination; dividing the scene parameter space into parameter subspaces; based on the parameter subspace, according to a target model and a scene parameter safety threshold, constructing a target function; and generating a safety analysis result according to the target function, so as to solve the problem that the reliability of the analysis result is relatively low because the safety analysis of the automatic driving system cannot be carried out based on the key parameters in the current safety analysis method of the automatic driving system.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method and system for analyzing the safety of an autonomous driving system based on scenario parameters. Background Art

[0002] With the continuous development of artificial intelligence theory and methods, autonomous driving technology has gradually matured and been applied. Since the behavior of the autonomous driving system is closely related to the scenario in which it operates, different parameters of the scenario will cause the autonomous driving system to exhibit different behaviors in it.

[0003] In order to ensure the safety of the autonomous driving system when in use, the safety analysis of the autonomous driving system based on simulation scenario parameters is currently used, aiming to evaluate the safety of the autonomous driving system in the scenario parameter space by simulating the operation of the autonomous driving system. By obtaining the scenario parameters of the vehicle in the automatic scene, such as rainfall, light intensity, road width, etc., the autonomous driving system is simulated in the simulator to analyze the safety of the autonomous driving system in the scenario parameter space.

[0004] However, the current safety analysis method for autonomous driving systems is based on a unified analysis of all scenario parameters, and is unable to obtain key parameters (i.e., parameters that have the greatest impact on safety) from the numerous scenario parameters. Since each scenario parameter has a different degree of impact on the autonomous driving system, it is impossible to conduct a safety analysis of the autonomous driving system based on key parameters, resulting in low credibility of the analysis results. Summary of the invention

[0005] The present application provides a scenario-based parameter-based autonomous driving system safety analysis method and system to solve the technical problem that the existing autonomous driving system safety analysis method is unable to perform safety analysis on the autonomous driving system based on key parameters, resulting in low credibility of the analysis results.

[0006] In a first aspect, the present application provides a method for analyzing the safety of an autonomous driving system based on scenario parameters, which is applied to an autonomous driving system of a vehicle; the method comprises: Acquire scene parameter data; the scene parameter data includes: a scene parameter vector, a scene parameter space, and a scene parameter safety threshold; the scene parameter vector is composed of a number of scene parameters of the vehicle in an autonomous driving scene, and the scene parameter space is a collection of scene parameter ranges; Based on the scene parameter data, a parameter vector instance and a feature quantity of the autonomous driving system under the parameter vector instance are obtained; the parameter vector instance includes: the scene parameter vector and the corresponding scene parameter range; Using the parameter vector instance and the feature quantity to train the linear model to obtain a target model; Inputting constraint conditions into the target model, the target model sorts the parameter vector instances and the feature quantities according to the Shapley values ​​from large to small according to the constraint conditions, determines the parameter vector instances and the feature quantities that are ranked first by a preset number, and obtains the target parameters; Combining any two of the target parameters to obtain a target parameter combination; Dividing the scene parameter space into a second preset number of parameter subspaces according to the target parameter combination; Based on the parameter subspace, constructing an objective function according to the target model and the scene parameter safety threshold; According to the objective function, a safety analysis result is generated.

[0007] In some embodiments, the step of obtaining a parameter vector instance and a feature quantity of the autonomous driving system located under the parameter vector instance based on the scene parameter data includes: Using a uniform distribution method, obtaining a third preset number of parameter vector instances in the scene parameter space; Inputting the parameter vector instance into the simulator to construct a scene instance corresponding to the parameter vector instance; The simulator is used to run the autonomous driving system, so that the autonomous driving system is placed in the scenario instance for simulation training, and characteristic quantities of the autonomous driving system are obtained.

[0008] In some embodiments, the step of sorting the parameter vector instances and the feature quantities according to the constraint conditions from large to small according to the Shapley values, determining the parameter vector instances and the feature quantities that are ranked first by a first preset number, and obtaining the target parameters comprises: According to the constraint conditions, the Shapley value of the parameter vector instance and the feature quantity is calculated using the SHAP method; Sort the Shapley values ​​from large to small to determine a first preset number of target Shapley values ​​that are ranked first; The parameter vector instance and the characteristic quantity corresponding to the target Shapley value are determined to obtain the target parameter.

[0009] In some embodiments, the target parameter combination includes: a first target parameter and a second target parameter; the step of dividing the scene parameter space into a second preset number of parameter subspaces according to the target parameter combination includes: Determine a scene parameter range of the target parameter combination in the scene parameter space; Based on the scene parameter range, dividing the first target parameter and the second target parameter according to a fourth preset number to obtain a first parameter subspace and a second parameter subspace; The first parameter subspace and the second parameter subspace are combined to obtain a second preset number of parameter subspaces; the second preset number is: N = L × L; Wherein, L is the fourth preset number.

[0010] In some embodiments, the target model is configured to: The parameter vector instance and the feature quantity are fitted to generate a target feature quantity.

[0011] In some embodiments, the objective function is: ; In the formula, δ ij For the i , j Quantitative security indicators of parameter subspaces; T is the scene parameter safety threshold; r ij Indicates i , j parameter subspaces; f ( X ) is the target feature.

[0012] In some embodiments, the step of generating a safety analysis result according to the objective function includes: According to the objective function, a quantitative safety index of the parameter subspace is calculated using a mixed integer linear programming method; Generate safety analysis results based on quantitative safety indicators.

[0013] In some embodiments, the step of generating a safety analysis result based on the quantitative safety index includes: If the quantitative safety index is less than or equal to 0, generating first information; If the quantitative security index is greater than 0, generating second information; The first information represents that the autonomous driving system is safe, and the second information represents that the autonomous driving system is unsafe.

[0014] In some embodiments, the method further comprises: Based on the security analysis result, a visualization result is generated; the visualization result is a heat map; the heat map is configured with a second preset number of squares, and the squares are used to display the first information or the second information.

[0015] The second aspect of the present application provides an autonomous driving system safety analysis system based on scenario parameters, comprising: Get module, configured as: Acquire scene parameter data; the scene parameter data includes: a scene parameter vector, a scene parameter space, and a scene parameter safety threshold; the scene parameter vector is composed of a number of scene parameters of the vehicle in an autonomous driving scene, and the scene parameter space is a collection of scene parameter ranges; Based on the scene parameter data, a parameter vector instance and a feature quantity of the autonomous driving system under the parameter vector instance are obtained; the parameter vector instance includes: the scene parameter vector and the corresponding scene parameter range; The training module is configured as follows: Using the parameter vector instance and the feature quantity to train the linear model to obtain a target model; The analysis result determination module is configured as follows: Inputting constraint conditions into the target model, the target model sorts the parameter vector instances and the feature quantities according to the Shapley values ​​from large to small according to the constraint conditions, determines the parameter vector instances and the feature quantities that are ranked first by a preset number, and obtains the target parameters; Combining any two of the target parameters to obtain a target parameter combination; Dividing the scene parameter space into a second preset number of parameter subspaces according to the target parameter combination; Based on the parameter subspace, constructing an objective function according to the target model and the scene parameter safety threshold; According to the objective function, a safety analysis result is generated.

[0016] The present application provides a method and system for analyzing the safety of an autonomous driving system based on scene parameters, which are applied to an autonomous driving system of a vehicle; the method comprises: obtaining scene parameter data; the scene parameter data comprises: a scene parameter vector, a scene parameter space, and a scene parameter safety threshold; the scene parameter vector is composed of a number of scene parameters of the vehicle in an autonomous driving scene, and the scene parameter space is a collection of the scene parameter ranges; based on the scene parameter data, obtaining a parameter vector instance and a feature quantity of the autonomous driving system under the parameter vector instance; the parameter vector instance comprises: the scene parameter vector and the corresponding scene parameter range; training a linear model using the parameter vector instance and the feature quantity to obtain a target model; inputting constraint conditions To the target model, the target model sorts the parameter vector instances and the feature quantities according to the Shapley values ​​from large to small according to the constraint conditions, determines the first preset number of the parameter vector instances and the feature quantities in the top order, and obtains the target parameters; combines any two parameters in the target parameters to obtain the target parameter combination; divides the scene parameter space into a second preset number of parameter subspaces according to the target parameter combination; based on the parameter subspace, constructs an objective function according to the target model and the scene parameter safety threshold; and generates a safety analysis result according to the objective function, so as to realize that the safety analysis method of the autonomous driving system can perform safety analysis on the autonomous driving system based on key parameters, thereby improving the credibility of the safety analysis result of the autonomous driving system. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solution of the present application, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0018] Figure 1 This is a flow chart of the safety analysis method of the autonomous driving system based on scenario parameters in this application; Figure 2 This is a process diagram of the safety analysis method of the autonomous driving system based on scenario parameters in this application; Figure 3 A method process diagram for the target model in this application; Figure 4 Schematic diagram of the visualization results in this application. DETAILED DESCRIPTION

[0019] In order to enable those skilled in the art to better understand the technical solutions in the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present application.

[0020] Since in some technologies, the safety analysis method of the autonomous driving system cannot perform safety analysis on the autonomous driving system based on key parameters, resulting in low credibility of the analysis results. In order to solve this technical problem, the present application provides a safety analysis method and system for an autonomous driving system based on scenario parameters. The safety analysis method and system for an autonomous driving system based on scenario parameters are described below: For example, with the continuous development of artificial intelligence theory and methods, autonomous driving technology has gradually matured and been applied. Since the behavior of the autonomous driving system is closely related to the scenario in which it operates, that is, different parameters of the scenario will cause the autonomous driving system to exhibit different behaviors in it. In order to ensure the safety of the autonomous driving system when in use, the current safety analysis of the autonomous driving system based on simulation scenario parameters is aimed at evaluating the safety of the autonomous driving system in the scenario parameter space by simulating the operation of the autonomous driving system. By obtaining the scenario parameters of the vehicle in the automatic scenario, such as rainfall, light intensity, road width, etc., the autonomous driving system is run in the simulator for simulation, so as to analyze the safety of the autonomous driving system in the scenario parameter space.

[0021] For example, the current test method for the automated driving system based on parameterized scenarios is as follows: Consider that a certain safety of the autonomous driving system can be represented by the constraint t≥T, where t is the characteristic quantity and T is the safety threshold (for example, for collision safety, the characteristic quantity is the minimum distance t between the vehicle and other objects under autonomous driving, and the threshold can be set to 0.1 meters, which constitutes the constraint t≥0.1).

[0022] Assume that a certain autonomous driving scenario consists of a parameter vector X=(x1, x2, ..., x m ) (such as rainfall, light intensity, road width, initial distance of other vehicles, maximum braking speed of other vehicles, etc.), and the ranges of the above parameters are [a1, b1], [a2, b2], ..., [a m , b m ], forming the parameter space R = [a1, b1] × [a2, b2] × ... × [a m , b mThen, we can randomly select K sets of values ​​from the parameter space R according to a certain probability distribution to form K sets of parameter vector instances (i.e., specific parameter vectors with values) X i (i=1, 2, ..., K). Each X i Then, for the above scenario instance, the autonomous driving system is run in the simulator for K simulations, and the feature quantity t is measured during the simulation. i (i=1, 2, ...., K) and judge t i ≥T is established, so as to analyze the safety of the autonomous driving system in the parameter space of this scenario.

[0023] For example, the current automated driving system testing method based on parameterized scenarios has the following problems: 1. Unable to reflect the relationship between parameter space and safety level The results of the current automated driving system testing methods based on parameterized scenarios are a series of parameter vectors and whether the automated driving system is safe under the parameters. These results can only show whether the system violates safety constraints under these discrete specific parameters. They lack quantification of the degree of safety and cannot effectively reflect the impact of different parameter values ​​on the degree of safety of the system.

[0024] 2. Lack of safety analysis of key parameters Different scenario parameters have different degrees of impact on the autonomous driving system. The current autonomous driving system testing method based on parameterized scenarios regards all parameters as equal and is unable to obtain key parameters (i.e., parameters with the greatest impact on safety) and perform safety analysis on the key parameters.

[0025] 3. Discreteness and locality of testing The current parameterized scenario-based autonomous driving system testing method directly uses randomly sampled scenario parameters to construct scenarios for simulation testing. These parameters are discrete sample points distributed in the parameter space. Therefore, the test results cannot effectively reflect the safety of the autonomous driving system in the entire parameter space.

[0026] like Figure 1 Shown is a flow chart of the method for analyzing the safety of an autonomous driving system based on scenario parameters in this application.

[0027] In a first aspect, the present application provides a method for analyzing the safety of an autonomous driving system based on scenario parameters, which is applied to an autonomous driving system of a vehicle; the method comprises the following steps: like Figure 2 Shown is a process diagram of the safety analysis method of the autonomous driving system based on scenario parameters in this application.

[0028] S100: Acquire scene parameter data; the scene parameter data includes: scene parameter vector X=(x1, x2, ..., x m ) (e.g., rainfall, light intensity, road width, initial distance of other vehicles, maximum braking speed of other vehicles, etc.), scene parameter space, the ranges of the parameters in the above scene parameter vector are [a1, b1], [a2, b2], ..., [a m , b m ], forming the parameter space R = [a1, b1] × [a2, b2] × ... × [a m , b m ], scene parameter safety threshold; the scene parameter vector is composed of several scene parameters of the vehicle in the autonomous driving scenario, and the scene parameter space is a set of scene parameter ranges.

[0029] S200: Based on the scene parameter data, obtain a parameter vector instance and a feature quantity of the autonomous driving system under the parameter vector instance; the parameter vector instance includes: the scene parameter vector and the corresponding scene parameter range.

[0030] The step of acquiring a parameter vector instance and a feature quantity of the autonomous driving system located under the parameter vector instance based on the scene parameter data comprises the following sub-steps: S210: Using a uniform distribution method, obtain a third preset number of parameter vector instances in the scene parameter space; sample in the parameter space R using a certain probability distribution (uniform distribution, normal distribution method, etc.) to obtain K groups of parameter vector instances X1, X2, ..., X K The uniform distribution method or other probability distribution methods can make the parameter vector instances obtained by sampling more random, thereby paving the way for the subsequent construction of scene instances.

[0031] S220: Input the parameter vector instance into the simulator to construct a scene instance corresponding to the parameter vector instance; K Construct the corresponding scene instance in the simulator.

[0032] S230: Run the autonomous driving system using the simulator, so that the autonomous driving system is placed in the scenario instance for simulation training, and obtain the characteristic quantity of the autonomous driving system. Run the autonomous driving system through the simulator for K simulations, measure and record the corresponding characteristic quantities t1, t2, ..., t K .

[0033] S300: Using the parameter vector instance and the feature quantity to train the linear model to obtain a target model; using the data pairs (X1, t1), (X2, t2), ..., (X K , t K ) Train a linear or piecewise linear proxy model (such as an affine function, a neural network with a ReLU activation function, etc.) to fit the above K groups of data.

[0034] like Figure 3 Shown is the method process diagram of the target model in the application.

[0035] Exemplarily, the target model is configured to: fit the parameter vector instance and the feature quantity to generate the target feature quantity.

[0036] S400: Input constraint conditions to the target model, and the target model sorts the parameter vector instances and the feature quantities according to the Shapley values ​​from large to small according to the constraint conditions, and determines the first preset number of the parameter vector instances and the feature quantities to obtain the target parameters. The first preset number can be determined based on actual analysis. The constraint conditions are safety test conditions for the autonomous driving system, such as collision test, etc.

[0037] In this embodiment, the target model first obtains the required corresponding parameter vector instances and feature quantities through the constraint conditions, and then uses the SHAP method to calculate the Shapley values ​​of the above parameter vector instances and feature quantities, and finally sorts the parameter vector instances and feature quantities from large to small according to the size of the Shapley values, so as to determine several parameter vector instances and the feature quantities with larger Shapley values ​​as target parameters.

[0038] For example, the SHAP method is a method for explaining model prediction based on game theory, and the Shapley value (SHAP) value is used to measure the contribution of each feature to the model prediction result. Through the SHAP method, the parameter vector instances and feature quantities are sorted from large to small according to the Shapley value according to the constraint conditions, and the parameter vector instances and feature quantities with higher sorting represent the parameter vector instances and feature quantities with greater contribution to the prediction result of the target model.

[0039] The step of sorting the parameter vector instances and the feature quantities according to the constraint conditions from large to small according to the Shapley values, determining the first preset number of parameter vector instances and the feature quantities that are ranked first, and obtaining the target parameters comprises the following sub-steps: S410: According to the constraint conditions, the Shapley values ​​of the parameter vector instance and the feature quantity are calculated using the SHAP method; S420: The Shapley values ​​are sorted from large to small, and the first preset number of target Shapley values ​​are determined; S430: The parameter vector instance and the feature quantity corresponding to the target Shapley value are determined to obtain the target parameter. The standard SHAP method is used to calculate the Shapley value of the input vector f(X), i.e., the data pair, and the Shapley values ​​are sorted and the corresponding h data pairs with the largest Shapley values ​​are selected as key parameters.

[0040] S500: Combining any two parameters among the target parameters to obtain a target parameter combination; the target parameter combination includes: a first target parameter and a second target parameter; the first target parameter and the second target parameter are any two parameters among the target parameters.

[0041] The step of dividing the scene parameter space into a second preset number of parameter subspaces according to the target parameter combination comprises the following sub-steps: S510: Determine the scene parameter range of the target parameter combination in the scene parameter space; first determine the scene parameter ranges [a1, b1] and [a2, b2] of the first target parameter x1 and the second target parameter x2 in the entire scene parameter space.

[0042] S520: Based on the scene parameter range, the first target parameter and the second target parameter are divided according to a fourth preset number to obtain a first parameter subspace and a second parameter subspace; for the first target parameter x1 and the second target parameter x2, their ranges are divided into L subintervals, and the specific steps are as follows: For the first target parameter x1, its range [a1, b1] is divided into L subintervals, and the width of each subinterval is: ; For the second target parameter x2, its range [a2, b2] is divided into L subintervals, and the width of each subinterval is: .

[0043] S530: Combine the first parameter subspace and the second parameter subspace to obtain a second preset number of parameter subspaces; the second preset number is: N = L × L; Wherein, L is the fourth preset number.

[0044] Specifically, for each sub-interval combination of the first target parameter x1 and the second target parameter x2, a parameter subspace is generated. For the i-th sub-interval of the first target parameter x1 [a1+(i-1)Δ1, a1+i-Δ1] and the j-th sub-interval of the second target parameter x2 [a2+(j-1)Δ2, a2+j-Δ2], the generated parameter subspace is: [a1+(i-1)Δ1, a1+i-Δ1]×[a2+(j-1)Δ2, a2+j-Δ2]×[a3, a3]×...×[a m , b m ].

[0045] S600: Divide the scene parameter space into a second preset number of parameter subspaces according to the target parameter combination.

[0046] Specifically, suppose that any first target parameter x1 and second target parameter x2 are analyzed: the entire parameter space [a1, b1]×[a2, b2]×...×[a m , b m ] is divided into L×L=L according to the first target parameter x1 and the second target parameter x2 2 parameter subspace. ij represents the i, jth parameter subspace, i.e. r ij =[a1+(b1-a1)i / L, a1+(b1-a1)(i+1) / L]×[a2+(b2-a2)i / l, a2+(b2-a2)(i+1) / L]×[a3, b3]...×[a m , b m ], where i, j = 0, 1, ..., L-1.

[0047] S700: Based on the parameter subspace, according to the target model and the scene parameter safety threshold, construct an objective function; the objective function is: ; In the formula, δ ij For the i , j Quantitative security indicators of parameter subspaces; T is the scene parameter safety threshold; r ij Indicates i , j parameter subspaces; f ( X ) is the target feature. f ( X ) represents the target model.

[0048] S800: Generate a safety analysis result according to the objective function.

[0049] The step of generating a safety analysis result according to the objective function comprises the following sub-steps: S810: According to the objective function, using the mixed integer linear programming method, calculate the quantitative safety index of the parameter subspace; S820: Generate a safety analysis result based on the quantitative safety index. Based on the objective function constructed above, use the standard mixed integer linear programming method to solve δ ij The value of δ ij That is, for the i, jth parameter subspace r ij The quantitative safety index, δ ij ≤0 means safe, δ ij >0 means unsafe, and δ ij The bigger it is, the less safe it is.

[0050] The step of generating a safety analysis result according to the quantitative safety index comprises the following sub-steps: S821: If the quantitative safety index is less than or equal to 0, generate first information; S822: If the quantitative safety index is greater than 0, generate second information; wherein the first information represents that the autonomous driving system is safe, and the second information represents that the autonomous driving system is unsafe. The quantitative safety index corresponds to each parameter subspace, and the safety of the autonomous driving system can be judged by the number of the first information and the second information.

[0051] like Figure 4 Shown is a schematic diagram of the visualization results in this application.

[0052] S900: Based on the security analysis result, generate a visualization result; the visualization result is a heat map; the heat map is configured with a second preset number of squares, and the squares are used to display the first information or the second information. In order to make it easier for detection personnel to observe the security analysis results, the present application uses a heat map to display the security analysis results.

[0053] In this embodiment, a heat map is used to visualize the security analysis results, that is, the heat map includes L 2 The squares that correspond to the parameter subspace are divided equally, and the color of each square corresponds to the security index δ of the parameter subspace. ij , where the darker the color, the safer the autonomous driving system, and the brighter the color, the more dangerous the autonomous driving system.

[0054] This application provides a method for analyzing the safety of an autonomous driving system based on scenario parameters, and the specific implementation methods are as follows: For example, the scene parameter vector includes: fog density, precipitation, precipitation sediment, solar altitude angle, solar azimuth angle, humidity and wind intensity, initial speed, and trigger distance; the constraint condition is a collision safety test; the scene parameter safety threshold is 0.2m; according to the constraint conditions, three target parameters are determined through the target model, such as: FOG-DENS (fog density), (PREC-DEP) precipitation sediment, (SUN-ALT) solar altitude angle; any two of the target parameters are combined to obtain a target parameter combination, and the scene parameter space is divided into several parameter subspaces according to the target parameter combination, and the target function is constructed using the target model and the scene parameter safety threshold; finally, according to the target function, the safety analysis results are generated and displayed in the form of a heat map, such as Figure 4 shown.

[0055] Specifically, Figure 4 As shown in (a), it is the safety analysis result when fog density and precipitation sediment are used as the target parameter combination. Through the safety analysis result, the safety of fog density and precipitation sediment under specific parameters can be obtained; Figure 4 As shown in (b), it is the safety analysis result when fog density and sun altitude angle are used as the target parameter combination. The safety of fog density and sun altitude angle under specific parameters can be obtained through the safety analysis result; Figure 4 As shown in (c), it is the safety analysis result when precipitation sediment and solar altitude angle are used as the target parameter combination. The safety of precipitation sediment and solar altitude angle under specific parameters can be obtained through the safety analysis result.

[0056] In this embodiment, the safety of the autonomous driving system under the range of target parameters can be observed through the heat map, such as Figure 4 As shown, the horizontal and vertical axes of the heat map represent the ranges of the two target parameters respectively. The heat map can be used to observe the safety of the autonomous driving system under any range of the two target parameters. The darker the color, the safer the autonomous driving system, and the brighter the color, the more dangerous the autonomous driving system.

[0057] The present application provides a method for analyzing the safety of an autonomous driving system based on scenario parameters. By giving an autonomous driving system and a parameter vector X=(x1, x2, ..., x m ) and parameter space R = [a1, b1] × [a2, b2] × ... × [a m , b m ], safety constraint t≥T, h key parameters with the greatest impact on safety are obtained through analysis, and these h parameters are combined in pairs, and the entire parameter range R is divided into L based on two of the key parameters. 2 parameter subspace Rij And calculate the quantitative security index δ of each parameter subspace ij And visualize it.

[0058] The second aspect of the present application provides an autonomous driving system safety analysis system based on scenario parameters, comprising: Get module, configured as: Acquire scene parameter data; the scene parameter data includes: a scene parameter vector, a scene parameter space, and a scene parameter safety threshold; the scene parameter vector is composed of a number of scene parameters of the vehicle in an autonomous driving scene, and the scene parameter space is a collection of scene parameter ranges; Based on the scene parameter data, a parameter vector instance and a feature quantity of the autonomous driving system under the parameter vector instance are obtained; the parameter vector instance includes: the scene parameter vector and the corresponding scene parameter range; The training module is configured as follows: Using the parameter vector instance and the feature quantity to train the linear model to obtain a target model; The analysis result determination module is configured as follows: Inputting constraint conditions into the target model, the target model sorts the parameter vector instances and the feature quantities according to the Shapley values ​​from large to small according to the constraint conditions, determines the parameter vector instances and the feature quantities that are ranked first by a preset number, and obtains the target parameters; Combining any two of the target parameters to obtain a target parameter combination; Dividing the scene parameter space into a second preset number of parameter subspaces according to the target parameter combination; Based on the parameter subspace, constructing an objective function according to the target model and the scene parameter safety threshold; According to the objective function, a safety analysis result is generated.

[0059] It is worth noting that the effects of the above system embodiment during operation can refer to the effects of the above method embodiment, which will not be described in detail here.

[0060] The above specific implementation methods further explain in detail the purpose, technical solutions and beneficial effects of the embodiments of the present application. It should be understood that the above are only specific implementation methods of the embodiments of the present application and are not used to limit the protection scope of the embodiments of the present application. Any modifications, equivalent substitutions, improvements, etc. made on the basis of the technical solutions of the embodiments of the present application should be included in the protection scope of the embodiments of the present application.

Claims

1. A method for analyzing the safety of an autonomous driving system based on scenario parameters, applied to an autonomous driving system of a vehicle; characterized in that: include: Get scene parameter data; The scene parameter data includes: a scene parameter vector, a scene parameter space, and a scene parameter safety threshold; the scene parameter vector is composed of a number of scene parameters of the vehicle in an autonomous driving scene, and the scene parameter space is a collection of scene parameter ranges; Based on the scene parameter data, a parameter vector instance and a feature quantity of the autonomous driving system under the parameter vector instance are obtained; the parameter vector instance includes: the scene parameter vector and the corresponding scene parameter range; Using the parameter vector instance and the feature quantity to train the linear model to obtain a target model; Inputting constraint conditions into the target model, the target model sorts the parameter vector instances and the feature quantities according to the Shapley values ​​from large to small according to the constraint conditions, determines the parameter vector instances and the feature quantities that are ranked first by a preset number, and obtains the target parameters; Combining any two of the target parameters to obtain a target parameter combination; Dividing the scene parameter space into a second preset number of parameter subspaces according to the target parameter combination; Based on the parameter subspace, constructing an objective function according to the target model and the scene parameter safety threshold; According to the objective function, a safety analysis result is generated.

2. The method for analyzing the safety of an autonomous driving system based on scenario parameters according to claim 1, characterized in that: The step of acquiring a parameter vector instance and a feature quantity of the autonomous driving system located under the parameter vector instance based on the scene parameter data comprises: Using a uniform distribution method, obtaining a third preset number of parameter vector instances in the scene parameter space; Inputting the parameter vector instance into the simulator to construct a scene instance corresponding to the parameter vector instance; The simulator is used to run the autonomous driving system, so that the autonomous driving system is placed in the scenario instance for simulation training, and characteristic quantities of the autonomous driving system are obtained.

3. The method for analyzing the safety of an autonomous driving system based on scenario parameters according to claim 1, characterized in that: The step of sorting the parameter vector instances and the feature quantities from large to small according to the Shapley values ​​according to the constraint conditions, determining the parameter vector instances and the feature quantities that are ranked first by a first preset number, and obtaining the target parameters comprises: According to the constraint conditions, the Shapley value of the parameter vector instance and the feature quantity is calculated using the SHAP method; Sort the Shapley values ​​from large to small to determine a first preset number of target Shapley values ​​that are ranked first; The parameter vector instance and the characteristic quantity corresponding to the target Shapley value are determined to obtain the target parameter.

4. The method for analyzing the safety of an autonomous driving system based on scenario parameters according to claim 1, characterized in that: The target parameter combination includes: a first target parameter and a second target parameter; the step of dividing the scene parameter space into a second preset number of parameter subspaces according to the target parameter combination includes: Determine a scene parameter range of the target parameter combination in the scene parameter space; Based on the scene parameter range, dividing the first target parameter and the second target parameter according to a fourth preset number to obtain a first parameter subspace and a second parameter subspace; The first parameter subspace and the second parameter subspace are combined to obtain a second preset number of parameter subspaces; the second preset number is: N = L × L; Wherein, L is the fourth preset number.

5. The method for analyzing the safety of an autonomous driving system based on scenario parameters according to claim 4, characterized in that: The target model is configured as: The parameter vector instance and the feature quantity are fitted to generate a target feature quantity.

6. The method for analyzing the safety of an autonomous driving system based on scenario parameters according to claim 5, characterized in that: The objective function is: ； In the formula, δ ij For the i , j Quantitative security indicators of parameter subspaces; T is the scene parameter safety threshold; r ij Indicates i , j parameter subspaces; f ( X ) is the target feature.

7. The method for analyzing the safety of an autonomous driving system based on scenario parameters according to claim 6, characterized in that: The step of generating a safety analysis result according to the objective function comprises: According to the objective function, a quantitative safety index of the parameter subspace is calculated using a mixed integer linear programming method; Generate safety analysis results based on quantitative safety indicators.

8. The method for analyzing the safety of an autonomous driving system based on scenario parameters according to claim 7, characterized in that: The step of generating a safety analysis result according to the quantitative safety index comprises: If the quantitative safety index is less than or equal to 0, generating first information; If the quantitative security index is greater than 0, generating second information; The first information represents that the autonomous driving system is safe, and the second information represents that the autonomous driving system is unsafe.

9. The method for analyzing the safety of an autonomous driving system based on scenario parameters according to claim 8, characterized in that: The method further comprises: Based on the security analysis result, a visualization result is generated; the visualization result is a heat map; the heat map is configured with a second preset number of squares, and the squares are used to display the first information or the second information.

10. A safety analysis system for an autonomous driving system based on scenario parameters, characterized in that: include: Get module, configured as: Get scene parameter data; The scene parameter data includes: a scene parameter vector, a scene parameter space, and a scene parameter safety threshold; the scene parameter vector is composed of a number of scene parameters of the vehicle in an autonomous driving scene, and the scene parameter space is a collection of scene parameter ranges; Based on the scene parameter data, a parameter vector instance and a feature quantity of the autonomous driving system under the parameter vector instance are obtained; the parameter vector instance includes: the scene parameter vector and the corresponding scene parameter range; The training module is configured as follows: Using the parameter vector instance and the feature quantity to train the linear model to obtain a target model; The analysis result determination module is configured as follows: Inputting constraint conditions into the target model, the target model sorts the parameter vector instances and the feature quantities according to the Shapley values ​​from large to small according to the constraint conditions, determines the parameter vector instances and the feature quantities that are ranked first by a preset number, and obtains the target parameters; Combining any two of the target parameters to obtain a target parameter combination; Dividing the scene parameter space into a second preset number of parameter subspaces according to the target parameter combination; Based on the parameter subspace, constructing an objective function according to the target model and the scene parameter safety threshold; According to the objective function, a safety analysis result is generated.

Images