Vehicle modeling prediction method based on physical information neural network

Abstract

The invention provides a vehicle modeling prediction method based on a physical information neural network, and belongs to the field of vehicle state prediction, and the method comprises the steps: S1, enabling the vehicle sensor data at the first n moments and the vehicle state predicted by the physical information neural network to form a time sequence at the k + 1 moment, and taking the time sequence as the input of the physical information neural network at the k + 1 moment; and S2, transmitting the vehicle sensor data at the moment k into a vehicle state estimator to estimate a dynamic change parameter which can change when the vehicle dynamically moves, transmitting the dynamic change parameter into a physical constraint part of the physical information neural network, enabling the physical information neural network to output a vehicle state at the moment k + 1, and obtaining a model constant parameter. According to the invention, dynamic modeling can be carried out, and the accuracy of vehicle modeling and prediction is improved.

Description

Technical Field

[0001] The invention belongs to the field of vehicle state prediction, and in particular relates to a vehicle modeling prediction method based on a physical information neural network. Background Art

[0002] Accurate vehicle dynamics models are crucial in developing advanced vehicle control systems. These models are used in a wide range of applications, from vehicle stability control to the development of autonomous driving technology. Traditional approaches to vehicle dynamics modeling are generally divided into two categories: physical modeling and data-driven modeling. Physical modeling relies on detailed mechanical and dynamic principles to describe the motion behavior of the vehicle. These models can range from simple single-track models to complex multi-body system models. Although physical models can provide deep mechanistic understanding and high prediction accuracy, they face the problem of difficult parameter identification and adjustment in practical applications, especially in long-term operation when vehicle parameters may change over time and usage. In addition, the establishment and validation of physical models usually require a large amount of experimental data and high costs, limiting their practicality in rapidly evolving vehicle technologies. Data-driven models capture the dynamic behavior of the vehicle by analyzing historical data and do not rely on the physical parameters of the vehicle. Such models can quickly adapt and predict the performance of the vehicle under different operating conditions without detailed physical understanding. However, data-driven models often lack physical interpretability and may not provide accurate predictions when dealing with unseen operating conditions. Summary of the invention

[0003] The purpose of the present invention is to propose a vehicle modeling and prediction method based on physical information neural network, which can perform dynamic modeling and improve the accuracy of vehicle modeling and prediction.

[0004] The present invention is achieved through the following technical solutions:

[0005] The vehicle modeling prediction method based on physical information neural network includes the following steps:

[0006] Step S1, at time k+1, the vehicle sensor data of the previous n times and the vehicle state predicted by the physical information neural network are formed into a time series as the input of the physical information neural network at time k+1;

[0007] Step S2, the vehicle sensor data at time k is transmitted to the vehicle state estimator to estimate the dynamic change parameters that will change during the dynamic motion of the vehicle, and the dynamic change parameters are transmitted to the physical constraint part of the physical information neural network, so that the physical information neural network outputs the vehicle state at time k+1, and obtains the model constant parameters, wherein the physical constraints include the 14-degree-of-freedom vehicle model equation, the Pacejka tire model equation and the longitudinal dynamics model equation, and the model constant parameters are included in the physical constraints, k is an integer, and n is an integer.

[0008] Furthermore, in step S1, the time series is represented as {[U k-n X k-n ] T ,[U k-n+1 X k-n+1 ] T …[U k X k ] T},U k represents the vehicle sensor data at time k, X k Represents the vehicle state predicted by the physical information neural network at time k.

[0009] Furthermore, in step S2, the 14-DOF vehicle model equation is Among them, I x is the moment of inertia about the x-axis, I y is the moment of inertia about the y-axis, I z is the moment of inertia about the z-axis, ∑M xz is the roll moment generated by the vertical suspension force, ∑M xt is the roll moment generated by the wheel force and the body weight, ∑M yz is the pitching moment generated by the vertical suspension force, ΣM yt is the pitching moment generated by the wheel force and the body weight, ΣM ztl is the yaw moment generated by the longitudinal force of the wheel, ΣM zts is the yaw moment generated by the lateral force of the wheel, ρ is the sideslip angle of the vehicle, θ is the pitch angle of the vehicle, ψ is the yaw angle of the vehicle, m represents the vehicle mass, represents the longitudinal velocity, represents the longitudinal acceleration, represents the lateral velocity, represents the lateral acceleration, ρ represents the angular acceleration of the sideslip angle, δ represents the front wheel turning angle, and F xfl Indicates the longitudinal force on the left front wheel, F xfr Indicates the longitudinal force on the right front wheel, F xrl Indicates the longitudinal force on the left rear wheel, F xrr Indicates the longitudinal force on the right rear wheel, F yflIndicates the lateral force on the left front wheel, F yfr Indicates the lateral force on the right front wheel, F yrl Indicates the lateral force on the left rear wheel, F yrr represents the lateral force on the right rear wheel, represents a time derivative, represents the second derivative with respect to time.

[0010] Furthermore, in step S2, the Pacejka tire model equation is: Where Y(x) represents the longitudinal force, lateral force or self-aligning moment, D represents the peak factor, C represents the shape factor, B represents the stiffness factor, E represents the curvature factor, X represents the sideslip angle or longitudinal slip rate, S represents the v is the vertical offset, S h is the horizontal offset.

[0011] Furthermore, in step S2, the longitudinal dynamics model equation is Among them, F x represents the longitudinal force of the vehicle, C m1 Indicates the force coefficient proportional to the driving torque, C m2 It represents the decrease coefficient of the effective conversion rate between torque and force as the vehicle speed increases, v x represents the longitudinal speed, T represents the driving torque applied to the wheel, and C r0 Denotes the rolling resistance coefficient, C d Represents the aerodynamic drag coefficient.

[0012] Furthermore, in step S2, the model constant parameter Ψ k Including I x ,I y ,I z , C d , D, C, B and E.

[0013] Furthermore, in step S2, the state estimator includes an unscented Kalman filter, a particle filter or an extended Kalman filter, and the dynamically changing parameter includes the mass m of the vehicle at the kth moment k , the distance from the center of mass of the vehicle to the front axle at the kth moment l r,k , the distance from the center of mass of the vehicle to the rear axle at the kth moment l f,k .

[0014] Furthermore, in step S2, when the dynamically changing parameters are transmitted to the physical constraint part of the physical information neural network, they are aligned in time and space.

[0015] The present invention has the following beneficial effects:

[0016] 1. At time k+1, the present invention forms a time series form of the vehicle sensor data of the previous n moments and the vehicle state predicted by the physical information neural network as the input of the physical information neural network at time k+1, and then transmits the vehicle sensor data at time k to the vehicle state estimator to estimate the dynamic change parameters that will change when the vehicle is in dynamic motion, and transmits the dynamic change parameters to the physical constraint part of the physical information neural network, so that the physical information neural network outputs the vehicle state at time k+1, and obtains the model constant parameters. In this way, while realizing the prediction of the vehicle state at the next moment, dynamic modeling is performed, and the uncertain model constant parameters of the vehicle are inversely obtained, which not only improves the accuracy of state prediction, but also increases the modeling accuracy of the vehicle in dynamic motion, and also has interpretability. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The present invention is further described in detail below with reference to the accompanying drawings.

[0018] Figure 1 It is a flow chart of the present invention.

[0019] Figure 2 It is a structural schematic diagram of the present invention. DETAILED DESCRIPTION

[0020] like Figure 1 and Figure 2 As shown, the vehicle modeling prediction method based on physical information neural network includes the following steps:

[0021] Step S1, at time k+1, the vehicle sensor data of the previous n times and the vehicle state predicted by the physical information neural network are formed into a time series as the input of the physical information neural network at time k+1;

[0022] Specifically, the time series is represented as {[U k-n X k-n ] T ,[U k-n+1 X k-n+1 ] T …[U k X k ] T},U k represents the vehicle sensor data at time k, X k Represents the vehicle state predicted by the physical information neural network at time k.

[0023] Step S2, transmitting the vehicle sensor data at time k to the vehicle state estimator to estimate the dynamic change parameters that will change during the dynamic motion of the vehicle, and transmitting the dynamic change parameters to the physical constraint part of the physical information neural network, so that the physical information neural network outputs the vehicle state at time k+1, and obtains the model constant parameters, wherein the physical constraints include the 14-degree-of-freedom vehicle model equation, the Pacejka tire model equation and the longitudinal dynamics model equation, and the model constant parameters are included in the physical constraints, k is an integer, and n is an integer;

[0024] Specifically, the state estimator includes an unscented Kalman filter, a particle filter or an extended Kalman filter, and the dynamically changing parameter Ψ u Including the mass m of the vehicle at the kth moment k , the distance l from the center of mass of the vehicle to the front axle at the kth moment r,k , the distance l from the center of mass of the vehicle to the rear axle at the kth moment f,k When the dynamically changing parameters are passed into the physical constraint part of the physical information neural network, they are aligned in time and space. In time, the data can be aligned by unified clock, interpolation or delay correction method, and in space, the data can be unified to the vehicle coordinate system by coordinate transformation, sensor calibration and fusion algorithm. This process is the prior art.

[0025] After the time series at time k+1 is input into the physical information neural network, the output after function fitting of the hidden layer of the physical information neural network is obtained. By substituting the output and the dynamic change parameters into the physical constraints, the vehicle state prediction at time k+1 can be obtained.

[0026] In the physical constraints, the 14-DOF vehicle model equation is Among them, I x is the moment of inertia about the x-axis, I y is the moment of inertia about the y-axis, I z is the moment of inertia about the z-axis, ∑M xz is the roll moment generated by the vertical suspension force, ∑M xt is the roll moment generated by the wheel force and the body weight, ∑M yz is the pitching moment generated by the vertical suspension force, ∑M yt is the pitching moment generated by the wheel force and the body weight, ∑M ztl is the yaw moment generated by the longitudinal force of the wheel, ∑M zts is the yaw moment generated by the lateral force of the wheel, ρ is the sideslip angle of the vehicle, θ is the pitch angle of the vehicle, ψ is the yaw angle of the vehicle, m is the vehicle mass, x is the longitudinal displacement, represents the longitudinal velocity, represents the longitudinal acceleration, y represents the lateral displacement, represents the lateral velocity, represents the lateral acceleration, ρ represents the angular acceleration of the sideslip angle, δ represents the front wheel turning angle, and F xfl Indicates the longitudinal force on the left front wheel, F xfr Indicates the longitudinal force on the right front wheel, F xrl Indicates the longitudinal force on the left rear wheel, F xrr Indicates the longitudinal force on the right rear wheel, F yfl Indicates the lateral force on the left front wheel, F yfr Indicates the lateral force on the right front wheel, F yrl Indicates the lateral force on the left rear wheel, F yrr represents the lateral force on the right rear wheel, represents a derivative with respect to time, represents the second derivative with respect to time.

[0027] The Pacejka tire model equation is Where Y(x) represents the longitudinal force, lateral force or self-aligning moment, D represents the peak factor, C represents the shape factor, B represents the stiffness factor, E represents the curvature factor, X represents the sideslip angle or longitudinal slip rate, S represents the v is the vertical offset of the tire fitting curve, S h is the horizontal offset of the tire fitting curve. When X is the slip angle, under the influence of the working conditions, Y(x) is the lateral force or the self-aligning moment; when X is the longitudinal slip rate, Y(x) is the longitudinal force.

[0028] The longitudinal dynamic model equation is Among them, F x represents the longitudinal force of the vehicle, C m1 The force coefficient proportional to the driving torque, C m2 It indicates the decrease coefficient of the effective conversion rate between torque and force as the vehicle speed increases. represents the longitudinal speed, T represents the driving torque applied to the wheel, and C r0 Denotes the rolling resistance coefficient, C d Represents the aerodynamic drag coefficient.

[0029] Model constant parameter ψ k Including the moment of inertia I around the x-axis of the vehicle coordinate system which is difficult to measure x , the moment of inertia I around the y-axis of the vehicle coordinate system y and the moment of inertia I around the z-axis of the vehicle coordinate system z , and the expensive aerodynamic drag coefficient C d , peak factor D, shape factor C, stiffness factor B and curvature factor E.

[0030] The above description is only a preferred embodiment of the present invention, and therefore cannot be used to limit the scope of implementation of the present invention. That is, equivalent changes and modifications made according to the scope of the patent application of the present invention and the contents of the specification should still fall within the scope covered by the patent of the present invention.

Claims

1. A vehicle modeling prediction method based on physical information neural network, characterized by: The steps include: Step S1, at time k+1, the vehicle sensor data of the previous n times and the vehicle state predicted by the physical information neural network are formed into a time series as the input of the physical information neural network at time k+1; Step S2, the vehicle sensor data at time k is transmitted to the vehicle state estimator to estimate the dynamic change parameters that will change during the dynamic motion of the vehicle, and the dynamic change parameters are transmitted to the physical constraint part of the physical information neural network, so that the physical information neural network outputs the vehicle state at time k+1, and obtains the model constant parameters, wherein the physical constraints include the 14-degree-of-freedom vehicle model equation, the Pacejka tire model equation and the longitudinal dynamics model equation, and the model constant parameters are included in the physical constraints, k is an integer, and n is an integer.

2. The vehicle modeling prediction method based on physical information neural network according to claim 1 is characterized in that: In step S1, the time series is represented as {[U k-n X k-n ] T ,[U k-n+1 X k-n+1 ] T …[U k X k ] T },U k represents the vehicle sensor data at time k, X k Represents the vehicle state predicted by the physical information neural network at time k.

3. The vehicle modeling prediction method based on physical information neural network according to claim 1 is characterized in that: In step S2, the 14-DOF vehicle model equation is: Among them, I x is the moment of inertia about the x-axis, I y is the moment of inertia about the y-axis, I z is the moment of inertia about the z-axis, ∑M xz is the roll moment generated by the vertical suspension force, ∑M xt is the roll moment generated by the wheel force and the body weight, ∑M yz is the pitching moment generated by the vertical suspension force, ∑M yt is the pitching moment generated by the wheel force and the body weight, ∑M ztl is the yaw moment generated by the longitudinal force of the wheel, ∑M zts is the yaw moment generated by the lateral force of the wheel, ρ is the sideslip angle of the vehicle, θ is the pitch angle of the vehicle, Ψ is the yaw angle of the vehicle, m represents the vehicle mass, represents the longitudinal velocity, represents the longitudinal acceleration, represents the lateral velocity, represents the lateral acceleration, ρ represents the angular acceleration of the sideslip angle, δ represents the front wheel turning angle, and F xfl Indicates the longitudinal force on the left front wheel, F xfr Indicates the longitudinal force on the right front wheel, F xrl Indicates the longitudinal force on the left rear wheel, F xrr Indicates the longitudinal force on the right rear wheel, F yfl Indicates the lateral force on the left front wheel, F yfr Indicates the lateral force on the right front wheel, F yrl Indicates the lateral force on the left rear wheel, F yrr represents the lateral force on the right rear wheel, represents a derivative with respect to time, represents the second derivative with respect to time.

4. The vehicle modeling prediction method based on physical information neural network according to claim 3 is characterized by: In step S2, the Pacejka tire model equation is: Where Y(x) represents the longitudinal force, lateral force or self-aligning moment, D represents the peak factor, C represents the shape factor, B represents the stiffness factor, E represents the curvature factor, X represents the sideslip angle or longitudinal slip rate, S represents the v is the vertical offset, S h is the horizontal offset.

5. The vehicle modeling prediction method based on physical information neural network according to claim 4 is characterized in that: In step S2, the longitudinal dynamics model equation is: Among them, F x represents the longitudinal force of the vehicle, C m1 The force coefficient proportional to the driving torque, C m2 It represents the decrease coefficient of the effective conversion rate between torque and force as the vehicle speed increases, v x represents the longitudinal speed, T represents the driving torque applied to the wheel, and C r0 Denotes the rolling resistance coefficient, C d Represents the aerodynamic drag coefficient.

6. The vehicle modeling prediction method based on physical information neural network according to claim 5 is characterized by: In step S2, the model constant parameter Ψ k Including I x ,I y ,I z , C d , D, C, B and E.

7. The vehicle modeling prediction method based on physical information neural network according to any one of claims 1 to 6, characterized in that: In step S2, the state estimator includes an unscented Kalman filter, a particle filter or an extended Kalman filter, and the dynamically changing parameter includes the mass m of the vehicle at the kth moment. k , the distance l from the center of mass of the vehicle to the front axle at the kth moment r,k , the distance l from the center of mass of the vehicle to the rear axle at the kth moment f,k .

8. The vehicle modeling prediction method based on physical information neural network according to any one of claims 1 to 6, characterized in that: In step S2, when the dynamically changing parameters are transmitted to the physical constraint part of the physical information neural network, they are aligned in time and space.

Images