Machine learning based vehicle dynamics parameter estimation method

Abstract

The application discloses a kind of vehicle dynamics parameter estimation methods based on machine learning, mainly through analysis, the important parameters of vehicle dynamics involved in simulation test software are obtained, and prediction method is provided for unknown parameter therein, for the situation that some parameters are difficult to obtain in engineering, provide a solution approach.Specifically, using grouping orthogonal test method, based on the test of simulation software platform, the importance of vehicle dynamics parameters is determined, and the important unknown parameters of vehicle dynamics are predicted using fully connected neural network, and the accuracy of prediction is verified by comparing the real vehicle test results with simulation test.The application focuses on analyzing and predicting unknown parameters in view of the characteristics that vehicle dynamics parameters are various and each parameter has different influence on vehicle dynamic response, effectively solves the problem of vehicle dynamics parameter sensitivity of automatic driving function, especially for reliable estimation of unknown vehicle dynamics sensitive parameters.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent networked vehicle testing, and in particular to a vehicle dynamics parameter estimation method based on machine learning. BACKGROUND

[0002] Simulation testing is one of the important means of intelligent networked vehicle automatic driving function testing. For the fusion perception, decision planning, control execution and other algorithm modules of automatic driving, simulation testing method can quickly verify the accuracy and reliability of the algorithm, without the risk of collision in real vehicle testing, and it is easy to generate various test scenarios. Therefore, in the field of automatic driving testing, the proportion of simulation testing is becoming higher and higher. However, an important problem in simulation testing is the accuracy of vehicle dynamics. The same control strategy has a large difference on different cars, because the vehicle dynamics response is different.

[0003] At present, the simulation testing software platform used in the industry has developed a vehicle dynamics model, which can relatively well simulate the dynamic response of the vehicle, but the prerequisite is to accurately master the vehicle dynamics parameters. In theory, the vehicle dynamics parameters should be given by the automobile design department or actually measured by test, but due to various reasons, the simulation test personnel may be difficult to obtain some parameters of the vehicle dynamics model, which leads to the inability to carry out subsequent testing work. SUMMARY

[0004] In view of the above, the present application aims to provide a vehicle dynamics parameter estimation method based on machine learning to solve the aforementioned technical problems.

[0005] The technical solution adopted by the present application is as follows:

[0006] The present application provides a vehicle dynamics parameter estimation method based on machine learning, which includes:

[0007] The sensitivity of the vehicle dynamics parameters is analyzed by using orthogonal test method combined with several simulation tests;

[0008] The unknown parameters in the simulation test are determined, and it is judged whether the dynamics response of the unknown parameters is sensitive based on the sensitivity analysis conclusion;

[0009] The target unknown parameters determined to be sensitive are predicted by using machine learning algorithm;

[0010] The prediction results are simulated and verified in the simulation software.

[0011] In at least one possible implementation, the sensitivity of the vehicle dynamics parameters is analyzed by:

[0012] The vehicle dynamics parameters are grouped, and an orthogonal test table is designed;

[0013] Design test cases, and select the corresponding dynamic response indicators;

[0014] Based on the orthogonal test table, the parameters are grouped for orthogonal test;

[0015] According to the correlation test value obtained from the orthogonal test results;

[0016] According to the comparison relationship between the correlation test value and the predetermined standard, the parameters are classified into sensitive parameters and non-sensitive parameters.

[0017] In at least one possible implementation, the employing a machine learning algorithm to predict the target unknown parameter determined to be sensitive includes:

[0018] Design a simulation test scenario and determine a target dynamic response indicator;

[0019] According to the preset default value of the target unknown parameter, a plurality of test cases are constructed;

[0020] Each test case is simulated in the designed scenario in sequence, and a first dynamic response indicator test value obtained from the simulation test is recorded;

[0021] The first dynamic response indicator test value is taken as input, and the target unknown parameter is taken as output, and a machine learning algorithm is established and trained;

[0022] The actual value of the dynamic response indicator is obtained through real vehicle testing, and is taken as input of the trained machine learning algorithm, and the prediction result of the target unknown parameter is output by the machine learning algorithm.

[0023] In at least one possible implementation, the constructing a plurality of test cases includes:

[0024] Based on the preset default value of the target unknown parameter, a value interval is demarcated;

[0025] Using the value interval, each target unknown parameter is expanded to construct a large number of test cases.

[0026] In at least one possible implementation, the constructing a plurality of test cases further includes: when constructing the test cases, all known parameters of vehicle dynamics in the simulation test are set to known values.

[0027] In at least one possible implementation, the simulating and verifying the prediction result in the simulation software includes:

[0028] The prediction result is input into the simulation software and simulated according to the same scenario to obtain a second dynamic response indicator test value;

[0029] Compare the differences in dynamic response between the first dynamic response index test value and the second dynamic response index test value;

[0030] If the difference is below the predetermined deviation threshold, the prediction is considered successful.

[0031] In at least one of the possible implementations, the determination of whether the dynamic response of the unknown parameter is sensitive based on the sensitivity analysis results includes: if the unknown parameter is determined to be insensitive, then the preset default value of the unknown parameter is used in the prediction stage.

[0032] In at least one possible implementation, the sensitivity analysis of vehicle dynamics parameters includes: performing a joint dynamics sensitivity analysis on the parameters of multiple sets of vehicle dynamics models in different autonomous driving simulation software.

[0033] Compared with existing technologies, the main design concept of this invention lies in analyzing and deriving important vehicle dynamics parameters involved in simulation testing software, thus providing direction for subsequent experimental parameter measurements. In particular, it offers prediction methods for important unknown parameters, providing a solution for situations where certain parameters are difficult to obtain in engineering. Specifically, a grouped orthogonal experimental method is employed, using simulation software platform tests as a basis to determine the importance of vehicle dynamics parameters. A fully connected neural network is then used to predict important unknown parameters of vehicle dynamics, and the accuracy of the predictions is verified by comparing real vehicle test results with simulation tests. This invention addresses the characteristics of numerous vehicle dynamics parameters and the different impacts of each parameter on vehicle dynamic response, focusing on analyzing and predicting unknown parameters. It effectively solves the sensitivity problem of vehicle dynamics parameters in autonomous driving functions, especially providing reliable estimations for unknown sensitive vehicle dynamics parameters. Attached Figure Description

[0034] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described below with reference to the accompanying drawings, wherein:

[0035] Figure 1 A schematic diagram of a machine learning-based vehicle dynamics parameter estimation method provided in an embodiment of the present invention;

[0036] Figure 2 This is a flowchart illustrating the parameter sensitivity determination method provided in an embodiment of the present invention. Detailed Implementation

[0037] Embodiments of the present application are described below in detail, examples of which are shown in the drawings, wherein the same or similar reference numbers represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary only, and are used to explain the present application, but cannot be interpreted as a limitation of the present application.

[0038] The present application proposes an embodiment of a vehicle dynamics parameter estimation method based on machine learning, specifically as shown in Figure 1 , which includes:

[0039] Step S1, using orthogonal test method and combining several simulation tests, the sensitivity of vehicle dynamics parameters is analyzed;

[0040] Specifically, according to the vehicle dynamics model of the automatic driving simulation software adopted by the user, the parameter module division is determined, and the vehicle dynamics response evaluation index is selected; using the orthogonal test method, the vehicle dynamics parameter sensitivity analysis is carried out according to the module grouping, and the number of parameters in each group is preferably not more than 9, and the P value of correlation test is used as the basis to judge whether a certain parameter is sensitive to the vehicle dynamics response, that is, the importance and influence. Referring to Figure 2 a flowchart of a vehicle dynamics parameter sensitivity analysis method is shown, it should be pointed out that if the simulation software is unchanged, the vehicle dynamics model can also be unchanged, in other words, the step S1 only needs to be executed once under the premise that the user does not change the simulation software. It can be supplemented that the parameters of multiple groups (preferably two groups) of vehicle dynamics models of different simulation software can be jointly analyzed for dynamics sensitivity to improve efficiency.

[0041] Step S2, determine the unknown parameters in the simulation test, and based on the analysis conclusion, judge whether the unknown parameters are sensitive to the dynamics response;

[0042] According to the results of vehicle dynamics parameter sensitivity analysis, for the unknown parameters of a simulation test, first check whether they are sensitive to the dynamics response, if not, take their default value in the simulation software; if sensitive, execute the next step.

[0043] Step S3, using machine learning algorithm to predict the target unknown parameters determined to be sensitive;

[0044] In detail, the process of parameter prediction using a fully connected neural network as an example can be referred to as follows:

[0045] (1) Design the scene of simulation test, and determine the dynamics response index (which can correspond to the simulation test scene and index in step S1);

[0046] (2) Set all known parameters of vehicle dynamics as known values, and then construct multiple test cases and number them according to preset default values of the target unknown parameters;

[0047] (3) Simulate each test case in order of numbering, and record the corresponding first dynamic response index test value;

[0048] (4) Take the above first dynamic response index test value as input and the corresponding target unknown parameter as output to train the fully connected neural network. Understandably, during the training process, the network layers, the number of neurons in each layer, the activation function, the learning rate and other hyperparameters need to be tuned, which will not be described here.

[0049] (5) Obtain the dynamic response index actual value through real vehicle testing, and take it as the actual input of the trained fully connected neural network, and output the prediction result of the target unknown parameter from the fully connected neural network.

[0050] Step S4, simulate the above prediction result in simulation software.

[0051] Specifically, the predicted value can be input into the simulation software and simulated according to the same test scene to obtain the second dynamic response index test value. The difference in dynamic response between the first dynamic response index test value (based on initial simulation) and the second dynamic response index test value (based on actual measurement) is compared. If the difference is below the established deviation threshold, it is determined that the prediction is successful, and if the difference is large, the neural network needs to be re-tuned and the prediction needs to be repeated.

[0052] Based on the above embodiment, the following examples are provided for specific description and introduction.

[0053] In this example, a certain simulation test software platform is used for simulation test of automatic driving function. The vehicle dynamics model used by the platform contains more than one hundred parameters, which are grouped as follows: vehicle body parameters, fuel vehicle transmission system parameters, electric vehicle transmission system parameters, braking system parameters, steering system parameters, suspension system parameters, and tire system parameters. Since the vehicle simulated in this example is a fuel vehicle, the electric vehicle transmission system parameters are not considered for the time being.

[0054] First step: Vehicle dynamics parameter sensitivity analysis.

[0055] According to the above grouping, orthogonal test is carried out to analyze the sensitivity of vehicle dynamics parameters. In actual operation, the designed simulation test scene is that the vehicle accelerates from static to 20 m / s on a straight road, and then turns left by 90°. The average acceleration in the first stage and the average angular acceleration in the second stage are taken as the dynamic response indicators. Taking the correlation P value 0.05 as the threshold as the judgment basis, the analysis result shows that the main parameters to be considered in the simulation test include the overall vehicle mass, the mass center height, the mass center relative front axle offset, the distance between the vehicle head and the front axle, the wheelbase, the torsional stiffness, the pedal lever ratio, the suspension force offset, the spring stiffness, the main cylinder diameter, the maximum steering wheel angle, the effective rolling radius of the wheel, the rolling resistance coefficient, the tire mass and the like.

[0056] Second step: determining unknown parameters and their influence degree.

[0057] In this case, the unknown parameters of the vehicle include the torsional stiffness, the pedal lever ratio, the rolling resistance coefficient and the shock lever ratio. According to the investigation, the shock lever ratio has no significant influence on the dynamic response of the vehicle, so the default value 0.7 of the software is adopted (the known values of the remaining parameters except the above unknown parameters are adopted).

[0058] Third step: prediction process of the remaining three important unknown parameters, which can be referred to as follows.

[0059] (1) Design the simulation test scene: the same as the simulation scene and the dynamic response indicators in the first step.

[0060] (2) Change the torsional stiffness, the pedal lever ratio and the rolling resistance coefficient according to their system default values in the range of ±80%, and divide each parameter into 10 equal parts on average, to generate 1000 test cases.

[0061] (3) Input each test case into the simulation software for testing, and record the corresponding average acceleration and average angular acceleration values of the two dynamic indicators.

[0062] (4) Construct a fully connected neural network with an initial state of 20 layers, 10 neural node units in each layer, a Sigmoid activation function and a learning rate of 0.05. The input is X=[average acceleration, average angular acceleration], the output is Y=[torsional stiffness, pedal lever ratio, rolling resistance coefficient], and 900 groups of data are randomly selected as the training set and 100 groups of data are selected as the test set. After multiple training and parameter adjustment, the prediction accuracy rate of the test set reaches 95.7%, and the corresponding neural network has 28 layers, 7 nodes in each layer, a Sigmoid activation function and a learning rate of 0.08.

[0063] (5) The vehicle is tested in a closed field according to the same test scene, and the measured average acceleration of the vehicle in straight line driving is 4.83 m / s 2, the average angular acceleration is 0.51 deg / s 2 The twist stiffness is 83.6 N.m / deg, the pedal lever ratio is 2.84, and the rolling resistance coefficient is 0.017.

[0064] The fourth step is to input the predicted values of the three unknown parameters into the simulation software, and test according to the same scene, and the average acceleration obtained is 4.97 m / s 2 , the average angular acceleration is 0.53 deg / s 2 It can be seen that, in this example, the errors are 2.9% and 3.9% respectively, which meet the established standard, and the prediction process is completed.

[0065] In summary, the main design concept of the application is to analyze the important parameters of vehicle dynamics involved in the simulation test software, to indicate the direction of subsequent test measurement parameters, especially to provide a prediction method for important unknown parameters, to provide a solution for the situation that some parameters are difficult to obtain in engineering. Specifically, the grouping orthogonal test method is used, the test of the simulation software platform is used as the basis to determine the importance of the vehicle dynamics parameters, the full connection neural network is used to predict the important unknown parameters of the vehicle dynamics, and the test results of the real vehicle are compared with the simulation test to verify the accuracy of the prediction. The application focuses on analyzing and predicting the unknown parameters among the various vehicle dynamics parameters, which effectively solves the problem of vehicle dynamics parameter sensitivity of automatic driving function, especially for reliable estimation of unknown vehicle dynamics sensitive parameters.

[0066] In the embodiment of the application, if the expression of the position is mentioned, it is based on the relative concept of the embodiment, and "at least one" means one or more, and "multiple" means two or more. "And / or" describes the association between the associated objects, which means that there can be three relationships, for example, A and / or B, which means that A exists alone, A and B exist simultaneously, and B exists alone. Wherein A, B can be singular or plural. The character " / " generally represents an "or" relationship between the associated objects before and after it. "At least one of the following" and similar expressions mean any combination of these items, including any combination of single or multiple items. For example, at least one of a, b and c can mean: a, b, c, a and b, a and c, b and c, or a and b and c, wherein a, b, c can be single or multiple.

[0067] The above detailed description of the embodiments shown in the drawings illustrates the structure, features and effects of the present application, but the above is only a preferred embodiment of the present application, and it should be noted that the technical features involved in the above embodiments and preferred modes can be reasonably combined and matched into various equivalent schemes by those skilled in the art without departing from or changing the design idea and technical effects of the present application; therefore, the present application is not limited to the implementation range shown in the drawings, and any changes or modifications made in accordance with the concept of the present application, or equivalent embodiments with equivalent changes, shall be within the scope of protection of the present application.

Claims

1. A machine learning based vehicle dynamics parameter estimation method, characterized by, The method comprises the following steps: sensitivity analysis of vehicle dynamics parameters is performed by using orthogonal test method combined with several simulation tests; unknown parameters in simulation tests are determined, and whether the dynamics response of the unknown parameters is sensitive is judged based on the conclusion of the sensitivity analysis; machine learning algorithm is used to predict the target unknown parameters determined as sensitive, including: designing simulation test scenarios and determining target dynamics response indicators; a plurality of test cases are constructed according to preset default values of the target unknown parameters; each test case is simulated in the designed scenario in sequence, and the first dynamics response indicator test value obtained by the simulation test is recorded; the machine learning algorithm is established and trained by taking the first dynamics response indicator test value as input and the target unknown parameter as output; the actual value of the dynamics response indicator is obtained by real vehicle test, and is taken as the input of the trained machine learning algorithm, and the prediction result of the target unknown parameter is output by the machine learning algorithm; the prediction result is simulated and verified in the simulation software.

2. The machine learning based vehicle dynamics parameter estimation method of claim 1, wherein, The sensitivity analysis of vehicle dynamics parameters comprises: grouping vehicle dynamics parameters and designing an orthogonal test table; designing test cases and selecting corresponding dynamics response indicators; orthogonal test is performed on each parameter group based on the orthogonal test table; correlation test values are obtained according to the orthogonal test results; parameters are classified as sensitive parameters and non-sensitive parameters according to the comparison relationship between the correlation test values and the predetermined standard.

3. The machine learning based vehicle dynamics parameter estimation method of claim 1, wherein, The construction of a plurality of test cases comprises: based on the preset default values of the target unknown parameters, the value interval is demarcated; a large number of test cases are constructed by extending each target unknown parameter using the value interval.

4. The machine learning based vehicle dynamics parameter estimation method of claim 1, wherein, When constructing test cases, all known parameters of vehicle dynamics in simulation tests are set to known values.

5. The machine learning based vehicle dynamics parameter estimation method of claim 1, wherein, The simulation verification of the prediction result in the simulation software comprises: the prediction result is input into the simulation software and simulated according to the same scenario to obtain the second dynamics response indicator test value; the dynamics response difference between the first dynamics response indicator test value and the second dynamics response indicator test value is compared; if the difference is lower than the predetermined deviation threshold, it is determined that the prediction is successful.

6. The machine learning based vehicle dynamics parameter estimation method of claim 1, wherein, The judgment of whether the dynamics response of the unknown parameter is sensitive based on the conclusion of the sensitivity analysis comprises: if the unknown parameter is judged as non-sensitive, the preset default value of the unknown parameter is used in the prediction link.

7. The machine learning based vehicle dynamics parameter estimation method according to any one of claims 1 to 6, characterized in that, The sensitivity analysis of vehicle dynamics parameters comprises: joint dynamics sensitivity analysis of parameters of multiple groups of vehicle dynamics models in different automatic driving simulation software.

Images