A method for identifying lateral dynamic parameters of four-wheel steering intelligent vehicles based on time series prediction network

Abstract

The present invention provides a method for identifying the lateral dynamic parameters of a four-wheel steering intelligent vehicle based on a time series prediction network, comprising a four-wheel steering vehicle dynamics model and a vehicle dynamics parameter identification model based on the vehicle's historical state. The method obtains vehicle state information from real vehicle sensors or a virtual simulation driving platform, selects vehicle state quantities that are easily measured by sensors as observer data sources, and divides them into historical features and future states, wherein historical features serve as observer inputs and future states serve as training labels. Combined with the three-degree-of-freedom four-wheel steering vehicle dynamics theory, the front and rear wheel lateral stiffness and moment of inertia that are difficult to measure by sensors are identified. This method takes into account the time series characteristics of the vehicle state and rationally designs a time series prediction network to achieve the simultaneous identification of multiple lateral dynamic parameters of a four-wheel steering intelligent vehicle. At the same time, the method can be combined with existing control algorithms to achieve more accurate vehicle stability control, effectively improving the driving safety of intelligent vehicles.

Description

Technical Field

[0001] The invention relates to a lateral dynamics identification method for a four-wheel steering intelligent vehicle based on a time series prediction network, and belongs to the field of automobile driving control. Background Art

[0002] The rapid development of electric vehicles, autonomous driving, and the gradual maturation of driverless technology have driven higher demands on vehicle control stability and safety. Vehicle lateral dynamics parameters are a key indicator of vehicle stability. However, these parameters are often difficult to measure directly with sensors. Therefore, lateral dynamic parameter identification based on observer technology has become a low-cost and efficient solution.

[0003] Influenced by technological development and market competition, high-end configurations represented by four-wheel steering have gradually become popular. Reasonable control can make four-wheel steering vehicles have significantly better flexibility and stability than traditional front-wheel steering vehicles, but their more complex dynamic characteristics also put higher requirements on the identification of their dynamic parameters.

[0004] To solve the above problems, the present invention provides a method for identifying the lateral dynamic parameters of a four-wheel steering intelligent vehicle based on a time series prediction network. This method uses a time series prediction network with embedded physical knowledge to observe the vehicle's lateral dynamic parameters. It can perform real-time identification of the front and rear wheel lateral stiffness and moment of inertia under different working conditions, thereby achieving more precise vehicle stability control. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to design a lateral dynamics parameter observer for a four-wheel steering intelligent vehicle that can be used in most working conditions in complex driving environments without adding additional sensors.

[0006] In order to construct a lateral dynamics parameter observer for a four-wheel steering intelligent vehicle that meets the above requirements, the present invention adopts the following technical solutions:

[0007] The identification method includes the following specific steps:

[0008] Step 1: Build a vehicle status data acquisition system. Use a virtual simulation driving platform or real vehicle sensors to collect status information of four-wheel steering vehicles under different driving conditions at regular time steps. The obtained vehicle status data is compiled into a dataset based on the vehicle's historical characteristics and future status, and divided into a training set, a validation set, a correction set, and a test set.

[0009] Step 2: Build a time series prediction network for lateral dynamic parameters of four-wheel steering vehicles based on LSTM and attention mechanism;

[0010] Step 3: Establish a four-wheel steering vehicle dynamics model and combine the dynamics formula as prior physical knowledge with the above-mentioned time series prediction network to form a complete four-wheel steering intelligent vehicle lateral dynamics parameter observer framework;

[0011] Step 4: Train the time series prediction network using the training set, optimizing it using mean square error as the loss function. The three-degree-of-freedom vehicle dynamics model is a simplified model that differs from the real vehicle and Carsim vehicle models, resulting in a certain systematic error between the actual prediction results and the true values. A sixth-order polynomial fit is implemented using the correction set to obtain accurate vehicle lateral dynamics parameters.

[0012] Step 5: Combine the trained observer with the lateral and longitudinal MPC trajectory tracking controller of the four-wheel steering vehicle, and verify the performance of the observer under different working conditions through Carsim-Simulink co-simulation.

[0013] Furthermore, the method for identifying lateral dynamic parameters of a four-wheel steering intelligent vehicle in step 1 is characterized by:

[0014] The vehicle's driving state information specifically includes the vehicle's lateral position X, longitudinal position Y, heading angle Yaw, lateral speed v x , longitudinal velocity v y , yaw rate ω, front wheel turning angle δ f , rear wheel turning angle δ r , throttle opening throttle, brake master cylinder pressure break;

[0015] The virtual simulation driving platform includes various devices that can be used to collect vehicle driving status data. The driving simulator hardware used in this method is Logitech G29. The virtual scene and vehicle simulation platform are built using the Prescan-Carsim-Simulink joint simulation method.

[0016] The vehicle history characteristics include the lateral speed v x , longitudinal velocity v y , yaw rate ω, front wheel turning angle δ f , rear wheel turning angle δ r , longitudinal control amount thr, front wheel angle change Δδ f , rear wheel angle change Δδ r , longitudinal control amount change Δthr, where the longitudinal control amount, front wheel angle change, rear wheel angle change, and longitudinal control amount change are obtained by performing simple data processing on the original state data:

[0017]

[0018] The maximum pressure of the brake master cylinder is break max When the vehicle is braking:

[0019]

[0020] When the vehicle is not braking:

[0021] thr t =throttle t

[0022] Then we have:

[0023] Δthr t =th r -thr t-1

[0024] The future state includes the lateral velocity v x , longitudinal velocity v y , yaw angular velocity ω.

[0025] Furthermore, the input of the time series prediction network in step 2 is consistent with the observer input, and the output is the lateral dynamic parameters and the front wheel longitudinal force F at each moment in the future state sequence of the four-wheel steering vehicle. x,f and rear wheel longitudinal force F x，r , where the lateral dynamic parameters include the front wheel cornering stiffness C f , rear wheel cornering stiffness C r 、Vehicle moment of inertia I z , front wheel side slip angle correction S f , rear wheel slip angle correction S r .

[0026] Furthermore, the four-wheel steering vehicle dynamics formula in step 3 includes the front and rear wheel slip angle calculation formula, the front and rear wheel lateral force calculation formula, the vehicle lateral and longitudinal acceleration, the yaw angular acceleration calculation formula, the front and rear wheel slip angle α f , α r The calculation formula is as follows:

[0027]

[0028] Among them, l f 、l r is the distance from the vehicle's center of mass to the front and rear axles.

[0029] Front and rear wheel lateral force F y,f 、F y,r The calculation formula is as follows:

[0030] F y,f =α f C f

[0031] F y，r =α r C r

[0032] Vehicle lateral and longitudinal acceleration d x d y , yaw angular acceleration d w The calculation formula is as follows:

[0033]

[0034] Among them, m is the mass of the vehicle.

[0035] Finally, the vehicle's lateral and longitudinal velocities and yaw angular velocity at time t+n are calculated using the following formula:

[0036]

[0037]

[0038] Among them, d t is the historical data sampling interval.

[0039] Furthermore, the loss function in step 4 is expressed as:

[0040]

[0041] Among them, n is the length of the time series, m is the dimension of the future vehicle state, and w j is the weight of different vehicle states, y ij is the true value of the future vehicle state, is the predicted value of the future vehicle state;

[0042] The training set includes various free driving conditions, and the correction set is the driving data of the vehicle performing uniform circular motion on a road with a constant adhesion coefficient, the same lane curvature, and a uniform gradient velocity distribution. The purpose is to eliminate factors other than vehicle speed and extract the numerical relationship between system error and vehicle speed through experimental methods.

[0043] The sixth-order polynomial for fitting the network output to the true value is expressed as:

[0044] C f =(a f v 6 +b f v 5 +c f v 4 +d f v 3 +e f v 2 +f f v+gf )C f.out

[0045] C r =(a r v 6 +b r v 5 +c r v 4 +d r v 3 +e r v 2 +f r v+g r )C r.out

[0046] Among them, C f.out 、C r.out is the front and rear wheel cornering stiffness output by the time series prediction network, C f 、C r is the front and rear wheel cornering stiffness actually output by the observer.

[0047] Furthermore, the control variables of the transverse and longitudinal MPC trajectory tracking controller in step 5 are the front and rear wheel steering angles and the desired acceleration a of the vehicle, and the steering angles of the four wheels are obtained through the underlying four-wheel steering angle distribution module. The throttle opening throttle and brake master cylinder pressure break are obtained by looking up the table in the bottom longitudinal control module, thereby achieving vehicle control.

[0048] The vehicle lateral dynamics parameter C in the controller is monitored by the observer f 、C r , I z Real-time updates are performed, and the vehicle's instability state is identified through the change amplitude of the lateral dynamic parameters, thereby achieving real-time updates of the vehicle's reference speed, improving the vehicle's stability and driving safety during path tracking.

[0049] Compared with the prior art, the present invention provides a method for identifying lateral dynamic parameters of a four-wheel steering intelligent vehicle based on a time series prediction network, which has the following beneficial effects:

[0050] The introduction of a time-series prediction network enables the present invention's lateral dynamics parameter observer to continuously learn. As data accumulates during vehicle operation, the observer can cover a wider range of operating conditions, providing reliable lateral dynamics parameter information for the vehicle control algorithm. In addition to fulfilling the functions of a traditional observer, this observer can also provide a certain degree of prediction of the vehicle's future lateral dynamics parameters. Although the prediction is only at the second level, it is sufficient to compensate for control lags caused by signal transmission time, algorithm execution time, and actuator delays, providing the control algorithm with dynamic parameters that are closer to the current moment. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a flow chart of a method for identifying lateral dynamic parameters of a four-wheel steering intelligent vehicle based on a time series prediction network proposed by the present invention;

[0052] Figure 2 is a schematic diagram of the three-degree-of-freedom four-wheel steering vehicle dynamics model in step 1 of this method;

[0053] Figure 3 This is a schematic diagram of the framework of the lateral dynamics parameter observer of the four-wheel steering intelligent vehicle in step three of this method;

[0054] Figure 4 This is the data collection scenario built by Prescan in the embodiment of this method;

[0055] Figure 5 is the Carsim vehicle front and rear wheel cornering stiffness of the correction set in the embodiment of the method;

[0056] Figure 6 This is a diagram of the estimation results of the front and rear wheel cornering stiffness of the vehicle in the correction set directly output by the time series prediction network in the embodiment of the method;

[0057] Figure 7 This is a diagram showing the estimation results of the front and rear wheel cornering stiffness of the vehicle, which is a correction set of the actual output of the observer after fitting with a sixth-order polynomial in the embodiment of the present method;

[0058] Figure 8 is the front and rear wheel cornering stiffness of the Carsim vehicle in the test set in the embodiment of this method;

[0059] Figure 9 is a diagram of the vehicle front and rear wheel cornering stiffness estimation results actually output by the observer in this method embodiment; DETAILED DESCRIPTION

[0060] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0061] Figure 1 FIG. 1 is a flow chart of a method for identifying lateral dynamic parameters of a four-wheel steering intelligent vehicle based on a time series prediction network proposed by the present invention. The method includes the following steps:

[0062] Step 1: Build a vehicle status data acquisition system. Use a virtual simulation driving platform or real vehicle sensors to collect status information of four-wheel steering vehicles under different driving conditions at regular time steps. The obtained vehicle status data is compiled into a dataset based on the vehicle's historical characteristics and future status, and divided into a training set, a validation set, a correction set, and a test set.

[0063] Step 2: Build a time series prediction network for lateral dynamic parameters of four-wheel steering vehicles based on LSTM and attention mechanism;

[0064] Step 3: Establish a four-wheel steering vehicle dynamics model and combine the dynamics formula as prior physical knowledge with the above-mentioned time series prediction network to form a complete four-wheel steering intelligent vehicle lateral dynamics parameter observer framework;

[0065] Step 4: Train the time series prediction network using the training set, optimizing it using mean square error as the loss function. The three-degree-of-freedom vehicle dynamics model is a simplified model that differs from the real vehicle and Carsim vehicle models, resulting in a certain systematic error between the actual prediction results and the true values. A sixth-order polynomial fit is implemented using the correction set to obtain accurate vehicle lateral dynamics parameters.

[0066] Step 5: Combine the trained observer with the lateral and longitudinal MPC trajectory tracking controller of the four-wheel steering vehicle, and verify the performance of the observer under different working conditions through Carsim-Simulink co-simulation.

[0067] Figure 1 The vehicle's driving status information in step 1 specifically includes the vehicle's lateral position X, longitudinal position Y, heading angle Yaw, lateral speed v x , longitudinal velocity v y , yaw rate ω, front wheel turning angle δ f , rear wheel turning angle δ r , throttle opening throttle, brake master cylinder pressure break;

[0068] The virtual simulation driving platform includes various devices that can be used to collect vehicle driving status data. The driving simulator hardware used in this method is Logitech G29. The virtual scene and vehicle simulation platform are built using the Prescan-Carsim-Simulink joint simulation method.

[0069] The vehicle history characteristics include the lateral speed v x , longitudinal velocity v y , yaw rate ω, front wheel turning angle δ f , rear wheel turning angle δ r , longitudinal control amount thr, front wheel angle change Δδ f , rear wheel angle change Δδ r , longitudinal control amount change Δthr, where the longitudinal control amount, front wheel angle change, rear wheel angle change, and longitudinal control amount change are obtained by performing simple data processing on the original state data:

[0070]

[0071] The maximum pressure of the brake master cylinder is break max When the vehicle is braking:

[0072]

[0073] When the vehicle is not braking:

[0074] thr t =throttle t

[0075] Then we have:

[0076] Δthr t =thr t -thr t-1

[0077] The future state includes the lateral velocity v x , longitudinal velocity v y , yaw angular velocity ω.

[0078] Figure 1 The input of the time series prediction network in step 2 is consistent with the observer input, and the output is the lateral dynamic parameters and the front wheel longitudinal force F at each moment in the future state sequence of the four-wheel steering vehicle. x，f and rear wheel longitudinal force F x，r , where the lateral dynamic parameters include the front wheel cornering stiffness C f , rear wheel cornering stiffness C r 、Vehicle moment of inertia I z , front wheel side slip angle correction S f , rear wheel slip angle correction S r .

[0079] Figure 1 The four-wheel steering vehicle dynamics formula in step 3 includes the front and rear wheel slip angle calculation formula, the front and rear wheel lateral force calculation formula, the vehicle's lateral and longitudinal acceleration, the yaw angular acceleration calculation formula, the front and rear wheel slip angle α f , α t The calculation formula is as follows:

[0080]

[0081] Among them, l f 、l r is the distance from the vehicle's center of mass to the front and rear axles.

[0082] Front and rear wheel lateral force F y，f 、F y，r The calculation formula is as follows:

[0083] F y,f =α f Cf

[0084] F y，r =α r C r

[0085] Vehicle lateral and longitudinal acceleration d x d y , yaw angular acceleration d w The calculation formula is as follows:

[0086]

[0087] Among them, m is the mass of the vehicle.

[0088] Finally, the vehicle's lateral and longitudinal velocities and yaw angular velocity at time t+n are calculated using the following formula:

[0089]

[0090] Among them, d t is the historical data sampling interval.

[0091] Figure 1 The loss function in step 4 is expressed as:

[0092]

[0093] Among them, n is the length of the time series, m is the dimension of the future vehicle state, and w j is the weight of different vehicle states, y ij is the true value of the future vehicle state, is the predicted value of the future vehicle state;

[0094] The training set includes various free driving conditions, and the correction set is the driving data of the vehicle performing uniform circular motion on a road with a constant adhesion coefficient, the same lane curvature, and a uniform gradient velocity distribution. The purpose is to eliminate factors other than vehicle speed and extract the numerical relationship between system error and vehicle speed through experimental methods.

[0095] The sixth-order polynomial for fitting the network output to the true value is expressed as:

[0096] C f =(a f v 6 +b f v 5 +c f v 4 +d f v 3 +e f v 2 +f f v+gf )C f.out

[0097] C r =(a r v 6 +b r v 5 +c r v 4 +d r v 3 +e r v 2 +f r v+g r )C r.out

[0098] Among them, C f.out 、C r.out is the front and rear wheel cornering stiffness output by the time series prediction network, C f 、C r is the front and rear wheel cornering stiffness actually output by the observer.

[0099] Figure 1 In step 5, the control variables of the lateral and longitudinal MPC trajectory tracking controller are the front and rear wheel angles and the desired acceleration a of the vehicle. The four wheel angles are obtained through the bottom four-wheel angle distribution module. The throttle opening throttle and brake master cylinder pressure break are obtained by looking up the table in the bottom longitudinal control module, thereby achieving vehicle control.

[0100] The vehicle lateral dynamics parameter C in the controller is monitored by the observer f 、C r , I z Real-time updates are performed, and the vehicle's instability state is identified through the change amplitude of the lateral dynamic parameters, thereby achieving real-time updates of the vehicle's reference speed, improving the vehicle's stability and driving safety during path tracking.

[0101] The specific operation process of this embodiment is as follows:

[0102] First, determine several important parameters of the vehicle to be controlled. In this embodiment, the vehicle to be controlled is a front-wheel drive four-wheel steering electric car with a mass of m = 1412 kg and a moment of inertia of I z =1109.7kg·m 2 , the distance from the center of mass to the front axle l f =1.015m, distance from center of mass to rear axle l r =1.895m, maximum torque of the front wheel drive motor T max =309.87N·m, maximum brake pressure of the brake master cylinderbreak max =9Mpa;

[0103] The above vehicle parameters are synchronously configured into the virtual simulation driving platform Carsim vehicle model, the observer vehicle physical parameters, the trajectory tracking control simulation Carsim vehicle model, and the four-wheel steering vehicle transverse and longitudinal MPC trajectory tracking controller vehicle physical parameters, and the controlled vehicle throttle and brake calibration table is prepared for the underlying vehicle longitudinal controller;

[0104] Use Prescan to build Figure 4 The driving road shown includes various working conditions such as straight roads, curves, roundabouts, and intersections. The vehicle driving data of experienced drivers on the road surfaces with adhesion coefficients of 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, and 0.9 are collected through the virtual simulation driving platform and made into a data set. The sampling interval is d t = 0.02s. The dataset is divided into a training set and a validation set to train the time series prediction network;

[0105] The virtual simulation driving platform is used to collect the driving data of vehicles driven by experienced drivers on a loop with a road adhesion coefficient of 0.9 and a radius of 100m at different speeds. The speeds are 20, 30, 40, 50, 60, 70, 80, 90, and 100 km / h. The collected data are made into a correction set, which is input into the trained time series prediction network to obtain Figure 6 The network output shown is calculated by the vehicle front and rear wheel slip angles and front and rear wheel lateral forces output by Carsim. Figure 5 The front and rear wheel cornering stiffness of the vehicle shown in the figure is fitted with the network output and Carsim results at different vehicle speeds using a sixth-order polynomial. The fitting results are shown in the figure below. Figure 7 As shown;

[0106] The driving data of vehicles driven freely by experienced drivers on a road with a road adhesion coefficient of 0.75 are collected through a virtual simulation driving platform and made into a test set. The test set is input into the observer to obtain Figure 9 The front and rear wheel cornering stiffness observation data shown are calculated by the front and rear wheel side slip angles and front and rear wheel lateral forces output by Carsim. Figure 8 The vehicle's front and rear wheel cornering stiffnesses are shown. The observer output is highly consistent with the CarSim data in terms of both value and trend. This completes the construction of the vehicle's lateral dynamics parameter observer.

[0107] Add the trained observer to the Simulink model and output the C f 、C r , I z The parameter list of the four-wheel steering vehicle lateral and longitudinal MPC trajectory tracking controller is updated at a frequency of 0.1s, and based on C f、C r For target vehicle speed v r Update in real time, the updated target speed v r The calculation formula is as follows:

[0108] v r =v ref -slog(w f (C f.risk -C f )+w r (C r.risk -C r ))

[0109] where v ref is the original target speed, C f.risk is the front wheel instability threshold, C r.risk is the rear wheel instability threshold, s is the target speed change adjustment weight, w f 、w r is the instability weight of the front and rear wheels;

[0110] The Carsim-Simulink joint simulation results show that the vehicle lateral stability control performance of the adaptive four-wheel steering vehicle lateral and longitudinal MPC trajectory tracking controller combined with the four-wheel steering intelligent vehicle lateral dynamics parameter identification method provided by the present invention is greatly improved.

[0111] The above embodiments are only part of the embodiments of the present invention. For ordinary technicians in this field, improvements, modifications and related applications can be made without departing from the technical principles of the present invention. These improvements should also be considered within the scope of protection of the present invention.

Claims

1. A method for identifying lateral dynamic parameters of a four-wheel steering intelligent vehicle, characterized in that: The following steps are involved: Step 1: Build a vehicle status data acquisition system. Use a virtual simulation driving platform or real vehicle sensors to collect status information of four-wheel steering vehicles under different driving conditions at regular time steps. The obtained vehicle status data is compiled into a dataset based on the vehicle's historical characteristics and future status, and divided into a training set, a validation set, a correction set, and a test set. Step 2: Build a time series prediction network for lateral dynamic parameters of four-wheel steering vehicles based on LSTM and attention mechanism; Step 3: Establish a four-wheel steering vehicle dynamics model and combine the dynamics formula as prior physical knowledge with the above-mentioned time series prediction network to form a complete four-wheel steering intelligent vehicle lateral dynamics parameter observer framework; Step 4: Train the time series prediction network using the training set and optimize it using mean square error as the loss function. The three-degree-of-freedom vehicle dynamics model is a simplified model and differs from the real vehicle and Carsim vehicle models, resulting in a certain systematic error between the actual prediction results and the true values. A sixth-order polynomial fit is implemented using the correction set to obtain accurate vehicle lateral dynamics parameters. Step 5: Combine the trained observer with the lateral and longitudinal MPC trajectory tracking controller of the four-wheel steering vehicle, and verify the performance of the observer under different working conditions through Carsim-Simulink co-simulation.

2. The method for identifying lateral dynamic parameters of a four-wheel steering intelligent vehicle according to claim 1, characterized in that: In step 1, the vehicle's driving status information specifically includes the vehicle's lateral position X, longitudinal position Y, heading angle Yaw, lateral speed v x , longitudinal velocity v y , yaw rate ω, front wheel turning angle δ f , rear wheel turning angle δ r , throttle opening throttle, brake master cylinder pressure break; The virtual simulation driving platform includes various devices that can be used to collect vehicle driving status data. The driving simulator hardware used in this method is Logitech G29. The virtual scene and vehicle simulation platform are built using the Prescan-Carsim-Simulink joint simulation method. The vehicle history characteristics include the lateral speed v x , longitudinal velocity v y , yaw rate ω, front wheel turning angle δ f , rear wheel turning angle δ r , longitudinal control amount thr, front wheel angle change Δδ f , rear wheel angle change Δδ r , longitudinal control amount change Δthr, where the longitudinal control amount, front wheel angle change, rear wheel angle change, and longitudinal control amount change are obtained by performing simple data processing on the original state data: The maximum pressure of the brake master cylinder is break max When the vehicle is braking: When the vehicle is not braking: thr t =throttle t Then we have: Δthr t =thr t -thr t-1 The future state includes the lateral velocity v x , longitudinal velocity v y , yaw angular velocity ω.

3. The method for identifying lateral dynamic parameters of a four-wheel steering intelligent vehicle according to claim 1, characterized in that: In the second step, the input of the time series prediction network is consistent with the observer input, and the output is the lateral dynamic parameters and the front wheel longitudinal force F at each moment in the future state sequence of the four-wheel steering vehicle. x，f and rear wheel longitudinal force F x，r , where the lateral dynamic parameters include the front wheel cornering stiffness C f , rear wheel cornering stiffness C r 、Vehicle moment of inertia I z , front wheel side slip angle correction S f , rear wheel slip angle correction S r .

4. The method for identifying lateral dynamic parameters of a four-wheel steering intelligent vehicle according to claim 1, characterized in that: In step 3, the four-wheel steering vehicle dynamics formula includes the front and rear wheel side slip angle calculation formula, the front and rear wheel lateral force calculation formula, the vehicle lateral and longitudinal acceleration, the yaw angular acceleration calculation formula, the front and rear wheel side slip angle α f , α r The calculation formula is as follows: Among them, l f 、l r is the distance from the vehicle's center of mass to the front and rear axles; Front and rear wheel lateral force F y，f 、F y，r The calculation formula is as follows: F y,f =a f C f F y，r =a r C r Vehicle lateral and longitudinal acceleration d x d y , yaw angular acceleration d w The calculation formula is as follows: Wherein, m is the mass of the vehicle; Finally, the vehicle's lateral and longitudinal velocities and yaw angular velocity at time t+n are calculated using the following formula: Among them, d t is the historical data sampling interval.

5. The method for identifying lateral dynamic parameters of a four-wheel steering intelligent vehicle according to claim 1, characterized in that: In step 4, the loss function is expressed as: Among them, n is the length of the time series, m is the dimension of the future vehicle state, and w j is the weight of different vehicle states, y ij is the true value of the future vehicle state, is the predicted value of the future vehicle state; The training set includes various free driving conditions, and the correction set is the driving data of the vehicle performing uniform circular motion on a road with a constant adhesion coefficient, the same lane curvature, and a uniform gradient velocity distribution. The purpose is to eliminate factors other than vehicle speed and extract the numerical relationship between system error and vehicle speed through experimental methods. The sixth-order polynomial for fitting the network output to the true value is expressed as: C f =(a f v 6 +b f v 5 +c f v 4 +d f v 3 +e f v 2 +f f v+g f )C f.out C r =(a r v 6 +b r v 5 +c r v 4 +d r v 3 +e r v 2 +f r v+g r )C r.out Among them, C f.out 、C r.out is the front and rear wheel cornering stiffness output by the time series prediction network, C f 、C r is the front and rear wheel cornering stiffness actually output by the observer.

6. The method for identifying lateral dynamic parameters of a four-wheel steering intelligent vehicle according to claim 1, characterized in that: In step 5, the control variables of the lateral and longitudinal MPC trajectory tracking controller are the front and rear wheel angles and the desired acceleration a of the vehicle, and the four wheel angles are obtained through the bottom four-wheel angle distribution module. The throttle opening throttle and brake master cylinder pressure break are obtained by looking up the table in the bottom longitudinal control module, thereby realizing vehicle control; The vehicle lateral dynamics parameter C in the controller is monitored by the observer f 、C r , I z Real-time updates are performed, and the vehicle's instability state is identified through the change amplitude of the lateral dynamic parameters, thereby achieving real-time updates of the vehicle's reference speed, improving the vehicle's stability and driving safety during path tracking.

Images