Vehicle kinetic parameter estimation method based on machine learning

Abstract

The invention discloses a vehicle dynamics parameter estimation method based on machine learning, which mainly obtains important parameters of vehicle dynamics involved in simulation test software through analysis, provides a prediction method for unknown parameters in the important parameters, and provides a solution for the condition that some parameters are difficult to obtain in engineering. Specifically, a grouping orthogonal test method is adopted, the test of a simulation software platform is taken as the basis, the importance of vehicle dynamics parameters is judged, a full-connection neural network is adopted to predict important unknown parameters of vehicle dynamics, and a real vehicle test result is compared with a simulation test to verify the prediction accuracy. The method focuses on analyzing and predicting unknown parameters in the vehicle dynamic parameters according to the characteristics that the types of the vehicle dynamic parameters are various and the influence of each parameter on the vehicle dynamic response is different, effectively solves the problem of sensitivity of the vehicle dynamic parameters with the automatic driving function, and particularly reliably estimates the unknown vehicle dynamic sensitive parameters.

Description

Technical Field

[0001] The present invention 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 Art

[0002] Simulation testing is one of the important means of testing the autonomous driving functions of intelligent connected vehicles. For the algorithm modules of autonomous driving, such as fusion perception, decision planning, and control execution, the simulation testing method can quickly verify the accuracy and reliability of the algorithm without the collision risk of real vehicle testing, and it is easy to generate a variety of test scenarios. Therefore, in the field of autonomous driving testing, the proportion of simulation testing is increasing. However, an important issue in simulation testing is the accuracy of vehicle dynamics. The same control strategy has a large difference in the impact on different vehicles because of different vehicle dynamic responses.

[0003] At present, the simulation test software platforms used in the industry have all developed vehicle dynamics models, which can simulate the dynamic response of the vehicle relatively well, but the premise is to accurately grasp the vehicle dynamics parameters. In theory, vehicle dynamics parameters should be given by the automobile design department or measured through actual experiments, but for various reasons, simulation testers may find it difficult to obtain some parameters of the vehicle dynamics model, resulting in the inability to carry out subsequent testing. Summary of the invention

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

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

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

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

[0008] Determine the unknown parameters in the simulation test, and judge whether the dynamic response of the unknown parameters is sensitive based on the sensitivity analysis conclusion;

[0009] Use machine learning algorithms to predict unknown parameters of targets identified as sensitive;

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

[0011] In at least one possible implementation, the analyzing the sensitivity of the vehicle dynamics parameter includes:

[0012] Group vehicle dynamics parameters and design orthogonal test tables;

[0013] Design test cases and select dynamic response indicators accordingly;

[0014] Based on the orthogonal test table, each parameter is grouped for orthogonal test;

[0015] Obtain correlation test value according to orthogonal test results;

[0016] Based on the comparison between the correlation test value and the established standard, the parameters are classified into sensitive parameters and non-sensitive parameters.

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

[0018] Design simulation test scenarios and determine target dynamic response indicators;

[0019] Constructing a plurality of test cases according to preset default values ​​of the target unknown parameters;

[0020] Simulate each test case in the designed scenario in sequence, and record the first dynamic response index test value obtained from the simulation test;

[0021] Using the first kinetic response index test value as input and the target unknown parameter as output, establishing and training a machine learning algorithm;

[0022] The actual value of the dynamic response index is obtained through real vehicle testing and used as the input of the trained machine learning algorithm, which then outputs the prediction results of the target unknown parameters.

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

[0024] Defining a value range based on a preset default value of the target unknown parameter;

[0025] By utilizing the value range, each of the target unknown parameters is expanded to construct a large number of test cases.

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

[0027] In at least one possible implementation manner, the performing simulation verification on the prediction result in the simulation software includes:

[0028] Input the prediction results into the simulation software and simulate according to the same scenario to obtain the test value of the second dynamic response index;

[0029] Comparing the kinetic response difference between the first kinetic response index test value and the second kinetic response index test value;

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

[0031] In at least one possible implementation, judging whether the dynamic response of the unknown parameter is sensitive based on the sensitivity analysis conclusion includes: if the unknown parameter is judged to be non-sensitive, using a preset default value of the unknown parameter in the prediction phase.

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

[0033] Compared with the prior art, the main design concept of the present invention is to derive the important parameters of vehicle dynamics involved in the simulation test software through analysis, to point out the direction for the subsequent test measured parameters, and especially to provide a prediction method for important unknown parameters, and to provide a solution for the situation where it is difficult to obtain certain parameters in engineering. Specifically, a group orthogonal test method is used, based on the test of the simulation software platform, to determine the importance of vehicle dynamics parameters, and a fully connected neural network is used to predict important unknown parameters of vehicle dynamics, and the actual vehicle test results are compared with the simulation test to verify the accuracy of the prediction. In view of the fact that there are many types of vehicle dynamics parameters and each parameter has a different impact on the vehicle's dynamic response, the present invention focuses on analyzing and predicting the unknown parameters, effectively solving the problem of vehicle dynamics parameter sensitivity of the autonomous driving function, and especially reliably estimates the unknown vehicle dynamics sensitive parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be further described below with reference to the accompanying drawings, in which:

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

[0036] Figure 2 A schematic diagram of a flow chart of a parameter sensitivity determination method provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0037] Embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be interpreted as limiting the present invention.

[0038] The present invention proposes an embodiment of a vehicle dynamics parameter estimation method based on machine learning. Specifically, Figure 1 shown, including:

[0039] Step S1, using an orthogonal test method combined with several simulation tests to analyze the sensitivity of vehicle dynamics parameters;

[0040] Specifically, according to the vehicle dynamics model of the autonomous driving simulation software used by the user, determine its parameter module division and select the vehicle dynamics response evaluation index; use the orthogonal test method to conduct vehicle dynamics parameter sensitivity analysis by module grouping, and the number of parameters in each group should preferably not exceed 9. The P value of the correlation test is used as the basis for judging whether a certain parameter is sensitive to the vehicle dynamics response, that is, the importance and influence. Figure 2 A flow chart of a vehicle dynamics parameter sensitivity analysis method is shown. It should be noted that if the simulation software does not change, the vehicle dynamics model may also remain unchanged. In other words, the user only needs to perform step S1 once without changing 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 combined for dynamic sensitivity analysis to improve efficiency.

[0041] Step S2, determining unknown parameters in the simulation test, and judging whether the dynamic response of the unknown parameters is sensitive based on the analysis conclusion;

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

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

[0044] To expand on this, the process of parameter prediction using a fully connected neural network as an example can be referenced as follows:

[0045] (1) Design a simulation test scenario and determine the dynamic response index (which may correspond to the simulation test scenario and index in step S1);

[0046] (2) setting all known parameters of vehicle dynamics to known values, and then constructing and numbering multiple test cases based on preset default values ​​of the target unknown parameters;

[0047] (3) Perform simulation tests on each test case in the order of their numbers and record the corresponding test values ​​of the first dynamic response index;

[0048] (4) Taking the above-mentioned first dynamic response index test value as input and the corresponding target unknown parameter as output, the fully connected neural network is trained. It can be understood that during the training process, the hyperparameters such as the number of network layers, the number of neurons in each layer, the activation function, and the learning rate need to be adjusted, which will not be elaborated in the present invention;

[0049] (5) The actual value of the dynamic response index is obtained through real vehicle testing and used as the actual input of the trained fully connected neural network, which then outputs the prediction result of the target unknown parameter.

[0050] Step S4: perform simulation verification on the above prediction results in simulation software.

[0051] Specifically, the predicted value can be input into the simulation software and the simulation can be performed according to the same test scenario to obtain the second kinetic response index test value, and the kinetic response difference between the first kinetic response index test value (based on the initial simulation) and the second kinetic response index test value (based on actual measurement) can be compared; if the difference is lower than the established deviation threshold, the prediction is determined to be successful; if the difference is large, the neural network needs to be re-adjusted and the prediction repeated.

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

[0053] In this example, a simulation test software platform is used to simulate the autonomous driving function. The vehicle dynamics model used by the platform contains more than 100 parameters, which are grouped as follows: body parameters, fuel vehicle transmission system parameters, electric vehicle transmission system parameters, brake 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] Step 1: Sensitivity analysis of vehicle dynamics parameters.

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

[0056] Step 2: Determine the unknown parameters and their influence.

[0057] The unknown parameters of the vehicle in this case include torsional stiffness, pedal lever ratio, rolling resistance coefficient, and shock absorber lever ratio. After investigation, it was found that the shock absorber lever ratio has no significant effect on the vehicle's dynamic response, so the software's default value of 0.7 is used (all known values ​​are used for the remaining known parameters except the above unknown parameters).

[0058] Step 3: The prediction process of the remaining three important unknown parameters can be referred to as follows.

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

[0060] (2) The torsional stiffness, pedal lever ratio, and rolling resistance coefficient were varied within the range of ±80% of their system default values. Each parameter was divided into 10 equal parts, generating a total of 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, which are two dynamic indicators.

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

[0063] (5) The vehicle was tested in a closed area using the same test scenario, and the average acceleration of the vehicle in a straight line was measured to be 4.83 m / s 2, the average angular acceleration is 0.51deg / s 2 Substituting the data into the neural network for prediction, the torsional stiffness was 83.6 N∙m / deg, the pedal lever ratio was 2.84, and the rolling resistance coefficient was 0.017.

[0064] Step 4: Input the predicted values ​​of the three unknown parameters into the simulation software and test it in the same scenario. The average acceleration is 4.97m / s 2 , the average angular acceleration is 0.53deg / s 2 ,It can be seen that in this example, the errors are 2.9% and 3.9% respectively, which meet the established standards, and the prediction process ends.

[0065] In summary, the main design concept of the present invention is to derive the important parameters of vehicle dynamics involved in the simulation test software through analysis, to point out the direction for the subsequent test measured parameters, and especially to provide a prediction method for important unknown parameters, and to provide a solution for the situation where it is difficult to obtain certain parameters in engineering. Specifically, a group orthogonal test method is used, based on the test of the simulation software platform, to determine the importance of vehicle dynamics parameters, and a fully connected neural network is used to predict the important unknown parameters of vehicle dynamics, and the actual vehicle test results are compared with the simulation test to verify the accuracy of the prediction. In view of the fact that there are many types of vehicle dynamics parameters and each parameter has a different impact on the vehicle's dynamic response, the present invention focuses on analyzing and predicting the unknown parameters, effectively solving the problem of vehicle dynamics parameter sensitivity of the automatic driving function, and especially reliably estimating the unknown vehicle dynamics sensitive parameters.

[0066] If the expressions expressing orientation are mentioned in the embodiments of the present invention, they are relative concepts based on the embodiments. In addition, "at least one" means one or more, and "plurality" means two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can be represented by: 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 describes in detail the structure, features and effects of the present invention based on the embodiments shown in the drawings, but the above is only a preferred embodiment of the present invention. It should be noted that the technical features involved in the above embodiments and their preferred methods can be reasonably combined and matched into a variety of equivalent schemes by those skilled in the art without departing from or changing the design ideas and technical effects of the present invention; therefore, the present invention is not limited to the scope of implementation shown in the drawings, and all changes made in accordance with the concept of the present invention, or modifications to equivalent embodiments with equivalent changes, which still do not exceed the spirit covered by the specification and drawings, should be within the protection scope of the present invention.

Claims

1. A vehicle dynamics parameter estimation method based on machine learning, characterized in that: include: The sensitivity of vehicle dynamics parameters is analyzed by using orthogonal test method combined with several simulation tests; Determine the unknown parameters in the simulation test, and judge whether the dynamic response of the unknown parameters is sensitive based on the sensitivity analysis conclusion; Use machine learning algorithms to predict unknown parameters of targets identified as sensitive; The prediction results are verified by simulation in the simulation software.

2. The vehicle dynamics parameter estimation method based on machine learning according to claim 1, characterized in that: The sensitivity analysis of vehicle dynamics parameters includes: Group vehicle dynamics parameters and design orthogonal test tables; Design test cases and select dynamic response indicators accordingly; Based on the orthogonal test table, each parameter is grouped for orthogonal test; Obtain correlation test value according to orthogonal test results; Based on the comparison between the correlation test value and the established standard, the parameters are classified into sensitive parameters and non-sensitive parameters.

3. The vehicle dynamics parameter estimation method based on machine learning according to claim 1, characterized in that: The use of a machine learning algorithm to predict the unknown parameters of a target determined to be sensitive includes: Design simulation test scenarios and determine target dynamic response indicators; Constructing a plurality of test cases according to preset default values ​​of the target unknown parameters; Simulate each test case in the designed scenario in sequence, and record the first dynamic response index test value obtained from the simulation test; Using the first kinetic response index test value as input and the target unknown parameter as output, establishing and training a machine learning algorithm; The actual value of the dynamic response index is obtained through real vehicle testing and used as the input of the trained machine learning algorithm, which then outputs the prediction results of the target unknown parameters.

4. The vehicle dynamics parameter estimation method based on machine learning according to claim 3 is characterized in that: The construction of multiple test cases includes: Defining a value range based on a preset default value of the target unknown parameter; By utilizing the value range, each of the target unknown parameters is expanded to construct a large number of test cases.

5. The vehicle dynamics parameter estimation method based on machine learning according to claim 3, characterized in that: The constructing of multiple test cases further includes: when constructing the test cases, all known parameters of vehicle dynamics in the simulation test are set to known values.

6. The vehicle dynamics parameter estimation method based on machine learning according to claim 1, characterized in that: The simulation verification of the prediction results in the simulation software includes: Input the prediction results into the simulation software and simulate according to the same scenario to obtain the test value of the second dynamic response index; Comparing the kinetic response difference between the first kinetic response index test value and the second kinetic response index test value; If the difference is below a predetermined deviation threshold, the prediction is considered successful.

7. The vehicle dynamics parameter estimation method based on machine learning according to claim 1, characterized in that: The method of judging whether the kinetic response of the unknown parameter is sensitive based on the sensitivity analysis conclusion includes: if the unknown parameter is judged to be non-sensitive, using a preset default value of the unknown parameter in the prediction phase.

8. The vehicle dynamics parameter estimation method based on machine learning according to any one of claims 1 to 7, characterized in that: The analysis of the sensitivity of vehicle dynamics parameters includes: performing a joint dynamics sensitivity analysis on parameters of multiple groups of vehicle dynamics models in different autonomous driving simulation software.

Images