Bench test device of vehicle intelligent system and acceleration safety assessment method thereof

Abstract

The invention discloses a rack testing device of a vehicle intelligent system and an acceleration safety evaluation method thereof, and belongs to the technical field of automobile safety, and the device comprises a vehicle testing cabinet, a wire control movable rack, a wire control steering rack, and a self-driving scene display system. According to the method, a tested system is iteratively fitted through a Gaussian process and a logistic function, and an agent model efficiently approaching the real tested system is generated to predict a security test result of the system; three acquisition functions are introduced to respectively explore the whole multi-dimensional continuous scene parameter space, namely, the safety performance of the system is represented by a small number of test scenes; accurately positioning a system security boundary; unsafe areas can be quickly identified, and dangerous scenes can be quickly found; and finally, giving the unsafe probability of the tested system in the test scene. The device and the method provided by the invention have relatively high test efficiency on the premise of ensuring the reliability and the accuracy of the vehicle intelligent system bench test.

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicle safety, and particularly relates to a bench test device for a vehicle intelligent system and an accelerated safety assessment method thereof. Background Art

[0002] Performing safety assessment on a vehicle intelligent system and defining a clear safety boundary are the prerequisites for its safe deployment and large-scale application.

[0003] Challenges in the safety assessment of vehicle intelligent systems: 1. Multidimensional continuity of interference models - as the number of interference variables increases, the fault search space increases exponentially; 2. Rare characteristics of potential faults - the non-convexity and unpredictability of related mathematical models make it difficult to capture them; 3. Introduction of a large number of neural networks - leading to the complexity of vehicle intelligent systems, and their inherent black-box non-interpretability makes it difficult for traditional stability analysis methods to comprehensively analyze safety performance.

[0004] When traditional bench test devices for vehicle intelligent systems (including the self-driving domain, chassis domain, and power domain) perform safety assessment, due to the emergence of the above three challenges, they cannot comprehensively and fully perform safety tests on the system under test in a short time, resulting in an increase in the cost of subsequent development and verification work such as real vehicle road tests, and thus affecting the R & D and application efficiency of vehicle intelligent systems. How to efficiently explore in the huge bench test scenario parameter space the scenarios that can trigger faults, i.e., insecurity, in the system under test, so as to significantly reduce the sample of scenario parameters required for testing, and at the same time maximize the credibility of the insecurity probability estimate, is the technical problem to be solved by the present invention. Summary of the Invention

[0005] To solve the above technical problems, the present invention adopts the following technical solutions:

[0006] A bench test device for a vehicle intelligent system, comprising: a vehicle test cabinet, an electronically controlled braking bench, an electronically controlled steering bench, and a self-driving scenario display system;

[0007] Among them, the vehicle test cabinet is used to load the dynamic model of the vehicle under test, the self-driving function and the chassis dynamic control function, run the management software and the accelerated test module of the test system, collect the analog signals of the electronically controlled braking bench and the electronically controlled steering bench, and send braking or steering commands to the electronically controlled braking bench and the electronically controlled steering bench respectively, and is responsible for the communication between the various parts of the bench test device for the vehicle intelligent system;

[0008] The electronically controlled braking bench is used to respond to the braking command of the vehicle in real time and generate the real braking force at the four wheels;

[0009] The steer-by-wire test bench includes a driver's seat, a pedal simulator, and a gearshift, which is used to respond to the steering commands of the vehicle's steering wheel in real time, generate the true angles at the four wheels, and conduct driver-in-the-loop tests or tests in the autonomous driving mode without human operation.

[0010] The autonomous driving scenario display system is used to run virtual animation software, map the operating state of the vehicle dynamics model obtained from the vehicle test cabinet in real time, and provide traffic flow feedback for real drivers or autonomous driving systems.

[0011] An accelerated safety assessment method for a bench test device of a vehicle intelligent system, which is used for the bench test device of the vehicle intelligent system, includes:

[0012] Step 1: Initialize the hyperparameters of the surrogate model of the Gaussian process for the vehicle intelligent system to be tested, and update the hyperparameters using maximum likelihood estimation based on the existing observations during the test process.

[0013] Step 2: Generate multiple test scenario parameters from the scenario parameter space, and use the surrogate model to predict the predicted output values of the vehicle intelligent system to be tested.

[0014] Step 3: Design three acquisition functions respectively for exploring the scenario parameter space, exploring the scenario safety boundary, and sampling in the unsafe or risky area. Based on the predicted output values and the acquisition functions, select test scenario parameters to test the vehicle intelligent system to be tested.

[0015] Step 4: After multiple iterations, conduct an efficient safety assessment of the vehicle intelligent system to be tested based on the observations, and estimate the unsafe probability of the vehicle intelligent system to be tested based on importance sampling.

[0016] An accelerated safety assessment device for a bench test device of a vehicle intelligent system, includes:

[0017] An initialization module, which initializes the hyperparameters of the surrogate model of the Gaussian process for the vehicle intelligent system to be tested, and updates the hyperparameters using maximum likelihood estimation based on the existing observations during the test process.

[0018] A prediction module, which generates multiple test scenario parameters from the scenario parameter space, and uses the surrogate model to predict the predicted output values of the vehicle intelligent system to be tested.

[0019] An acquisition function acquisition module, which designs three acquisition functions respectively for exploring the scenario parameter space, exploring the scenario safety boundary, and sampling in the unsafe or risky area. Based on the predicted output values and the acquisition functions, select test scenario parameters to test the vehicle intelligent system to be tested.

[0020] A safety assessment module, which conducts an efficient safety assessment of the vehicle intelligent system to be tested based on the observations after multiple iterations, and estimates the unsafe probability of the vehicle intelligent system to be tested based on importance sampling.

[0021] An electronic device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps of the accelerated safety assessment method of the bench test device for the vehicle intelligent system are implemented.

[0022] A non-transitory computer-readable storage medium stores a computer program thereon. When the computer program is executed by a processor, the steps of the accelerated safety assessment method of the bench test device for the vehicle intelligent system are implemented.

[0023] The present invention has the following beneficial effects:

[0024] 1. By establishing a Gaussian process surrogate model for the vehicle intelligent system under test and updating the parameters of the model using maximum likelihood estimation based on the existing observations in the test process, the present invention can quickly model the complex vehicle intelligent system under test with an unpredictable model containing multi-dimensional interferences.

[0025] 2. By designing three acquisition functions respectively for exploring the scenario parameter space, exploring the scenario safety boundary, and sampling in the unsafe or risky area, the present invention can reasonably select test scenario parameters for extensive and targeted exploration to evaluate the safety of the system under test, especially the safety of the system under test under rare faults.

[0026] 3. By transforming the safety assessment of the vehicle intelligent system under test into a Bayesian optimization problem and estimating the unsafe probability of the vehicle intelligent system under test in the test scenario based on importance sampling, the present invention can efficiently calculate the unsafe probability of the system under test under the intelligent vehicle internal control algorithm and complex external driving environment.

[0027] Compared with the existing test methods, the device and method proposed by the present invention have higher test efficiency on the premise of ensuring the credibility and accuracy of the bench test of the vehicle intelligent system. Description of the Drawings

[0028] Figure 1 is a structural diagram of the bench test device for the vehicle intelligent system of the present invention;

[0029] Figure 2 is a framework diagram of the accelerated safety assessment of the bench test device for the vehicle intelligent system of the present invention. Detailed Embodiments

[0030] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0031] In view of the complexity of large-scale neural networks, the black-box security assessment method of the present invention regards the vehicle intelligent system as a black-box system under test. Without in-depth understanding of the prior knowledge of the system, it widely explores the multi-dimensional continuous test scenario parameter space to identify and evaluate potential faults of the system. The Bayesian optimization algorithm constructs a surrogate model to approximate the objective function and intelligently selects the next evaluation point to find the optimal solution of the objective function within fewer iterations, thereby improving the security assessment efficiency. The importance sampling technique uses the cross-entropy optimization algorithm to solve the optimal proposal distribution, that is, the fault distribution of the system under test, to achieve differential sampling and weight calculation of samples, and improve the efficiency and accuracy of fault probability estimation.

[0032] For the bench test device of the vehicle intelligent system of the present invention, in order to improve the test evaluation efficiency on the basis of ensuring the test evaluation accuracy, the security assessment of the vehicle intelligent system is transformed into a Bayesian optimization problem. That is, the system under test is iteratively fitted through Gaussian processes and logistic functions to generate a surrogate model that efficiently approximates the real system under test to predict the security test results of the system; then three acquisition functions are introduced to explore the entire multi-dimensional continuous scenario parameter space respectively, that is, to characterize the security performance of the system with a small number of test scenarios; accurately locate the system security boundary; quickly identify the unsafe area, accelerate the discovery of dangerous scenarios; and finally give the probability of the system under test being unsafe in the test scenario.

[0033] As Figure 1 shown, the bench test device of the vehicle intelligent system of the present invention consists of four parts of hardware: a vehicle test cabinet, a wire-controlled braking bench, a wire-controlled steering bench, and a self-driving scenario display system.

[0034] The vehicle test cabinet is used to load the dynamic model of the vehicle under test, the self-driving function and the chassis dynamic control function, run the management software of the test system and the acceleration test module. In addition, it collects the analog signals (wheel braking force, wheel angle) of the wire-controlled braking bench and the wire-controlled steering bench, and sends braking or steering commands to the wire-controlled braking bench and the wire-controlled steering bench respectively, and is responsible for the communication between the various parts of the bench test device of the vehicle intelligent system.

[0035] The wire-controlled braking bench is used to respond in real time to braking commands such as the opening of the vehicle's brake pedal and generate the real braking force at the four wheels.

[0036] The steer-by-wire test bench includes a driver's seat, a pedal simulator, and a gearshift, which is used to respond in real time to the steering commands of the vehicle's steering wheel, generate the true angles at the four wheels, and conduct tests for driver-in-the-loop testing or autonomous driving mode without human operation.

[0037] The autonomous driving scenario display system is used to run virtual animation software, map in real time the operating state of the vehicle's dynamic model obtained from the vehicle test cabinet, and provide real-time feedback on road conditions such as traffic flow, such as pedestrians and vehicle distances, to real drivers or autonomous driving systems.

[0038] Typical test objects for the vehicle intelligent system bench include: trajectory planning algorithms under the interaction between the autonomous driving domain and the chassis domain, ACC / AEB (Adaptive Cruise Control / Automatic Emergency Braking) functions, and other driving assistance functions; vehicle handling stability and path tracking control algorithms under the coordinated control of multiple systems (cross-system) in the chassis domain; active and passive fault-tolerant control functions under chassis failure operation.

[0039] Typical test scenario parameters for the vehicle intelligent system include: vehicle speed, vehicle weight, accelerator pedal opening, brake pedal opening, steering wheel angle, relative position / speed with other vehicles / people / objects, road surface adhesion coefficient, etc., and these parameters are usually considered to be independent of each other.

[0040] The test scenario parameters are defined as:

[0041] , , ;

[0042] Among them, is the parameter space of the scenario, is the test scenario parameter array, which consists of n parameters, is the n-dimensional real number space, is the first test scenario parameter, and so on, is the nth test scenario parameter.

[0043] The probability distribution that the test scenario parameters follow is:

[0044] ;

[0045] Among them, is the probability distribution of the test scenario parameters.

[0046] The safety specification of the system under test is defined as , which is an important indicator for evaluating the safety performance of the system. When the state of the vehicle violates this specification, such as not being within the specified threshold, the system is unsafe; otherwise, the system is safe.

[0047] Introduce the black box mapping function As follows, map the given array of test scenario parameters to the output as follows:

[0048] ;

[0049] where y = f(x) is the black box mapping function, . When y = 1, the system violates the safety specification, and the at this time represents a risk scenario, and the test is considered unqualified; when y = 0, the system is safe, and the at this time represents a safe scenario, and the test is considered qualified.

[0050] The acceleration safety assessment method of the vehicle intelligent system of the present invention includes the following three objectives:

[0051] Objective 1: Achieve rapid falsification of the vehicle intelligent system under test, that is, solve the risk scenario that causes the system to violate the safety specification, as follows:

[0052] ;

[0053] Objective 2: Analyze and determine the risk scenario with the highest probability, that is, solve the risk scenario with the highest probability under the condition of the probability distribution of scenario parameters, as follows:

[0054] ;

[0055] Objective 3: Estimate the probability of the vehicle intelligent system under test being unsafe, that is, generate test scenario cases using the probability distribution of test scenario parameters, and test the vehicle intelligent system under test; then estimate the probability that the vehicle intelligent system under test violates the safety specification according to the test results, as follows:

[0056] ;

[0057] where is the probability that the system violates the safety specification, that is, the probability that the system is unsafe is the expected value under the probability distribution, is the number of tests that have been carried out.

[0058] In order to efficiently evaluate the safety of the vehicle intelligent system under test, the present invention uses a Gaussian process and a logistic function to establish a surrogate model of the vehicle intelligent system under test for predicting whether the vehicle intelligent system under test is safe, that is, whether the test is qualified.

[0059] As Figure 2 shown, the acceleration safety assessment method of the vehicle intelligent system of the present invention includes the following four steps:

[0060] Step 1: Initialize the hyperparameters of the Gaussian process surrogate model and update them using maximum likelihood estimation based on the existing observations in the testing process.

[0061] Step 2: Generate multiple test scenario parameters from the scenario parameter space and use the surrogate model to predict the output value of the intelligent system of the vehicle under test, i.e., the predicted output value (i.e., the test result).

[0062] Step 3: Design three acquisition functions for exploring the scenario parameter space, exploring the scenario safety boundary, and sampling in the unsafe (i.e., risky) area respectively. Select test scenario parameters based on the predicted output value and the acquisition functions to test the intelligent system of the vehicle under test.

[0063] Step 4: After multiple iterations, perform an efficient safety assessment on the intelligent system of the vehicle under test based on the observations, and estimate the unsafe probability of the intelligent system of the vehicle under test based on importance sampling.

[0064] Specifically, in Step 1, establish a surrogate model for the intelligent system of the vehicle under test to predict the output value and uncertainty of the system y = f(x) under the input of the test parameters. Based on the Gaussian process, combine the logit function and the logistic function to establish this surrogate model, and map the value range of from [0, 1] to the real number interval. The parameter is used to adjust the steepness of the logit function curve:

[0065] ;

[0066] Among them, is the parameter for adjusting the steepness of the logit function curve.

[0067] Construct and update the following Gaussian process (the surrogate model of the intelligent system of the vehicle under test) based on the observations:

[0068] ;

[0069] Among them, is the array of test scenario parameters from the 1st to the Nth time, = u is the mean function, and u is the average value of the test scenario parameter array. is the Gaussian process function; is the kernel function, which is related to the spatial information of the parameter . The kernel function adopts the squared exponential kernel function:

[0070] ;

[0071] Among them, and are any two sets of parameters within the and are the signal variance and length scale of the kernel function, respectively; is the natural exponential function.

[0072] The hyperparameters of this Gaussian process are: , based on the observed values , update the hyperparameters using maximum likelihood estimation:

[0073] ;

[0074] where, are the updated hyperparameters of the Gaussian process, is the -th test, and is the observed value under the parameter is the observed value from the 1st to the -th test; is the value of that maximizes .

[0075] Step 2, use the surrogate model to predict the output of the intelligent system of the vehicle under test, i.e., the test result. Generate test scenario parameters under a uniform distribution , and the logit function value predicted by the Gaussian process is denoted as . The unobserved and the tested value follow a joint Gaussian distribution. Use Bayes' formula to obtain the conditional probability distribution of :

[0076] ;

[0077] ;

[0078] ;

[0079] where, ) and are the specific expressions of the two parameters in the conditional probability distribution, is the mean function, is the kernel function, is the inverse of the kernel function.

[0080] Introduce the inverse function of the logit function, i.e., the logistic function h( ) to map the predicted value to [0, 1], and obtain the final predicted output value of the intelligent system of the vehicle under test:

[0081] ;

[0082] Among them, is the predicted output value; is the Gaussian process predicted value of the logistic function.

[0083] In summary, when the predicted value , it is considered that the test scenario parameter array at this time is a risk scenario and the test fails; when the predicted value , it is considered that the test scenario parameter array represents a safe scenario and the test passes.

[0084] In step 3, a collection function is designed to select the optimal sample points , , of the test to improve the safety assessment efficiency of the vehicle under - test system. There are three forms of the collection function:

[0085] Select the scenario parameter with the largest standard deviation of the Gaussian process prediction. At this time, the surrogate model lacks relevant knowledge of this parameter and needs to be explored to reduce the uncertainty of the model. Eventually, it can quickly cover the entire parameter space, as follows:

[0086] ;

[0087] Among them, is the first collection function, is the standard deviation of the Gaussian process prediction.

[0088] Select the safety - margin collection function, as follows:

[0089] ;

[0090] Among them, is the second collection function; the parameter is the adjustment parameter for safety - margin collection, used to coordinate the exploration and exploitation of the margin; t is the number of iterations of safety - margin collection; represents the preferred high - probability boundary parameter, and other boundaries are gradually explored as the iteration increases. The derivative of the logistic function is the analytical form of the scenario parameter boundary, that is, when , the corresponding test parameter is the boundary between the risk and safe scenarios. Among them, is the predicted output value.

[0091] Select the predicted risk scenario for sampling to estimate the probability that the system under test is unsafe, that is, the test fails, as follows:

[0092] ;

[0093] Among them, is the third acquisition function; the upper confidence bound of the Gaussian process prediction value is ; is an indicator function. When the upper bound is greater than 0.5, it is considered a risk scenario and sampled with probability ; otherwise, it is a safe scenario and no sampling will be performed.

[0094] In step 4, after N iterations of prediction and observation, according to the measured observation value the intelligent system of the vehicle under test is evaluated for safety, and the most likely risk scenario parameters and the unsafe probability of the intelligent system of the vehicle under test are obtained.

[0095] Importance sampling is used to obtain a sufficiently accurate probability estimate value with as few test iteration numbers as possible. After introducing the likelihood-weighted proposal distribution the probability that the system under test violates the safety specification, that is, the unsafe probability of the intelligent system of the vehicle under test is:

[0096] ;

[0097] Among them, is the unsafe probability, that is, the probability of failed test; is the parameter of the likelihood-weighted proposal distribution, is the expected value under the probability distribution, is the expected value under the q(x) probability distribution.

[0098] Based on the bench test device and the accelerated test method of the vehicle intelligent system, the present invention has high accuracy in estimating the unsafe probability of the system under test. At the same time, the number of test samples and the number of iterations used are small, and the test time consumption is short. It enables the tester to efficiently obtain the input parameters of the failed test, that is, the risk scenario, under the bench test of the intelligent system of the vehicle under test, so as to carry out targeted function improvement, guide the enhancement of the control performance of the vehicle intelligent system under weak working conditions, provide sufficient safety evaluation data and clear safety boundaries for subsequent vehicle tests, and shorten the cycle of product safety deployment and large-scale application.

[0099] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The solutions in the embodiments of the present invention can be implemented in various computer languages. For example, object-oriented programming languages such as Java and interpreted scripting languages such as JavaScript, etc.

[0100] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for realizing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0101] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that realizes the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0102] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide steps for realizing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0103] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications once they learn the basic creative concepts. Therefore, the appended claims are intended to be construed as including the preferred embodiments as well as all changes and modifications falling within the scope of the present invention.

[0104] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention also intends to include these modifications and variations.

Claims

1. A bench test device for a vehicle intelligent system, characterized in that Including: A vehicle test cabinet, a wire-controlled braking bench, a wire-controlled steering bench, and a self-driving scenario display system; Among them, the vehicle test cabinet is used to load the dynamic model of the vehicle under test, the self-driving function, and the chassis dynamic control function, run the management software and the acceleration test module of the test system, collect the analog signals of the wire-controlled braking bench and the wire-controlled steering bench, and send braking or steering commands to the wire-controlled braking bench and the wire-controlled steering bench respectively, and is responsible for the communication between the various parts of the bench test device of the vehicle intelligent system; The wire-controlled braking bench is used to respond to the braking command of the vehicle in real time and generate the real braking force at the four wheels; The wire-controlled steering bench includes a driver's seat, a pedal simulator, and a gearshift, and is used to respond to the steering command of the vehicle's steering wheel in real time, generate the real steering angles at the four wheels, and conduct driver-in-the-loop tests or tests in the self-driving mode of unmanned operation; The self-driving scenario display system is used to run virtual animation software, map the running state of the vehicle's own dynamic model obtained by the vehicle test cabinet in real time, and provide traffic flow feedback for real drivers or self-driving systems.

2. An acceleration safety assessment method for a bench test device of a vehicle intelligent system, which is used for the bench test device of the vehicle intelligent system as described in claim 1, characterized in that, Including: Step 1, initialize the hyperparameters of the surrogate model of the Gaussian process of the intelligent system of the vehicle under test, and update the hyperparameters using maximum likelihood estimation based on the existing observations in the test process; Step 2, generate multiple test scenario parameters from the scenario parameter space, and use the surrogate model to predict the predicted output values of the intelligent system of the vehicle under test; Step 3, design three acquisition functions respectively for exploring the scenario parameter space, exploring the scenario safety boundary, and sampling in the unsafe or risk area, and select test scenario parameters based on the predicted output values and the acquisition functions to test the intelligent system of the vehicle under test; Step 4, after multiple iterations, conduct an efficient safety assessment of the intelligent system of the vehicle under test based on the observations, and estimate the unsafe probability of the intelligent system of the vehicle under test based on importance sampling.

3. The accelerated safety assessment method for the bench test device of the vehicle intelligent system according to claim 2, wherein the test scenario parameters are defined as: ， ， ； Among them, is the parameter space of the scenario, is the test scenario parameter array, consisting of n parameters, is the n-dimensional real number space, is the first test scenario parameter, and so on, is the nth test scenario parameter; The probability distribution followed by the test scenario parameters is: ； Among them, is the probability distribution of the test scenario parameters; The safety specification of the system under test is defined as , when the vehicle's state violates this specification, such as not being within the specified threshold, the system is unsafe; otherwise the system is safe; Introduce the black-box mapping function As follows, map the given array of test scenario parameters to the output as follows: ； Among them, y = f(x) is the black box mapping function of ; when y = 1, the system violates the security specification, and the at this time represents a risk scenario, and it is considered that the test fails; when y = 0, the system is safe, and the at this time represents a safe scenario, and it is considered that the test passes.

4. The acceleration safety assessment method of the bench test device for the vehicle intelligent system according to claim 3, characterized in that In Step 1, a surrogate model of the intelligent system of the vehicle under test is established to predict the output value and uncertainty of the system y = f(x) under the input of test parameters; based on the Gaussian process, a surrogate model is established by combining the logit function and the logistic function, and the value range is mapped from [0, 1] to the real number interval: ； Among them, is a parameter for adjusting the steepness of the logit function curve; Construct and update the following Gaussian process based on the observations: ； Among them, is the parameter array of the 1st to the Nth test scenarios, =u is the mean function, and u is the average value of the test scenario parameter array, is the Gaussian process function; is the kernel function, which is related to the spatial information of the parameter The kernel function adopts the squared exponential kernel function: ； Among them, and are any two sets of parameters within the test, and are the signal variance and length scale of the kernel function respectively; is the natural exponential function; The hyperparameters of the Gaussian process are: , based on the observed values , update the hyperparameters using maximum likelihood estimation: ； Among them, is the hyperparameter after Gaussian process update, is the -th test, and is the observation value under parameter is the observation value from the 1st to the -th test; is the value that makes reach the maximum value.

5. The acceleration safety assessment method of the bench test device for the vehicle intelligent system according to claim 3, characterized in that, The test scenario parameters include: vehicle speed, vehicle weight, accelerator pedal opening, brake pedal opening, steering wheel angle, relative position / velocity with other vehicles / people / objects, and road surface adhesion coefficient, and the parameters are independent of each other.

6. The acceleration safety assessment method for the bench test device of the vehicle intelligent system according to claim 3, characterized in that Step 2, predicting the output value of the intelligent system of the vehicle under test by using the surrogate model, including: generating test scenario parameters under uniform distribution , and the logit function value predicted by the Gaussian process is denoted as ; the unobserved and the tested value obey the joint Gaussian distribution, and the conditional probability distribution of is obtained by using the Bayesian formula: ； ； ； Among them, ) and are the specific expressions of two parameters in the conditional probability distribution, is the mean function, is the kernel function, is the inverse of the kernel function; Introduce the logistic function h( ) to map the predicted value to [0, 1], obtaining the final predicted output value of the intelligent system of the vehicle under test : ； Among them, is the predicted output value; is the Gaussian process predicted value of the logistic function; When the predicted value is obtained, it is considered that the test scenario parameter array represents a risk scenario and the test fails; when the predicted value is obtained, it is considered that the test scenario parameter array represents a safe scenario and the test passes.

7. The acceleration safety assessment method of the bench test device for the vehicle intelligent system according to claim 3, characterized in that, Step 3 includes: designing a collection function and selecting the optimal sample points for testing , , ; Select the standard deviation of the Gaussian process prediction The maximum scenario parameter: ； Among them, is the first acquisition function, is the standard deviation of the Gaussian process prediction; Select the safety boundary acquisition function: ； Among them, is the second acquisition function; the parameter is the adjustment parameter for safe boundary acquisition, which is used to coordinate the exploration and exploitation of the boundary; t is the number of iterations of safe boundary acquisition; is the preferred high-probability boundary parameter; the derivative of the logistic function is the analytical form of the scenario parameter boundary. When the predicted output value is, the corresponding test parameter is the boundary of the risk and safety scenarios; Select the predicted risk scenarios for sampling, and estimate the probability that the system under test is unsafe or fails the test: ； Among them, the upper confidence bound of the Gaussian process prediction value is ; is an indicator function. When the upper bound is greater than 0.5, it is considered a risk scenario and sampled with probability ; otherwise, it is a safe scenario and not sampled.

8. The acceleration safety assessment method for the bench test device of the vehicle intelligent system according to claim 3, wherein After N iterations of prediction and observation, according to the measured observation values Perform a safety assessment on the intelligent system of the vehicle under test to obtain the parameters of the most likely risk scenario and the probability of insecurity of the intelligent system of the vehicle under test; Step 4 includes: Proposed distribution introducing likelihood weighting After that, the probability of insecurity of the intelligent system of the vehicle under test is as follows: ； Among them, is the probability of being unsafe, that is, the probability of failing the test; is the parameter of the likelihood weighted proposal distribution, is the expected value under the probability distribution, is the expected value under the q(x) probability distribution.

9. An acceleration safety evaluation device for a bench test device of a vehicle intelligent system, characterized in that, Including: An initialization module, which initializes the hyperparameters of the surrogate model of the Gaussian process of the intelligent system of the vehicle under test, and updates the hyperparameters using maximum likelihood estimation based on the existing observations in the test process; A prediction module, which generates multiple test scenario parameters from the scenario parameter space, and uses the surrogate model to predict the predicted output values of the intelligent system of the vehicle under test; An acquisition function acquisition module, which designs three acquisition functions respectively for exploring the scenario parameter space, exploring the scenario safety boundary, and sampling in the unsafe or risk area, and selects test scenario parameters based on the predicted output values and the acquisition functions to test the intelligent system of the vehicle under test; The safety assessment module performs efficient safety assessment on the intelligent system of the vehicle under test according to the observed values after multiple iterations, and estimates the unsafe probability of the intelligent system of the vehicle under test based on importance sampling.

10. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the accelerated safety assessment method of the bench test device for the vehicle intelligent system according to any one of claims 2 to 8.

11. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the accelerated safety assessment method of the bench test device for the vehicle intelligent system according to any one of claims 2 to 8.

Images