A Parameter Reconstruction Method for Trucks under Variable Loads Based on an RBF Optimal Ridge Observer

Abstract

The present invention discloses a parameter reconstruction method for a truck under variable load based on an RBF optimal ridge observer, and the steps are as follows: training an RBF neural network based on test data and the truck dynamics equation to obtain an RBF optimal ridge parameter neural network capable of calculating the optimal ridge parameters through sensor signals; estimating the front wheelbase, the height of the center of mass of the sprung mass, and the equivalent cornering stiffness of the tire based on the cross-iteration of the basis vectors; the present invention estimates the front wheelbase, the height of the center of mass of the sprung mass, and the cornering stiffness of the front and rear tires in real time based on the existing vehicle-mounted sensors, so as to provide relatively accurate model parameters for model reconstruction and vehicle control, and improve the control efficiency and accuracy.

Description

Technical Field

[0001] The invention belongs to the technical field of vehicle parameter estimation, and in particular relates to a truck parameter reconstruction method under variable load based on an RBF optimal ridge observer. Background Art

[0002] Trucks are an important carrier of transportation in my country, and they have a huge number of vehicles in China. The cargo transported by trucks is different each time. The difference in mass between empty and fully loaded can reach several tons to tens of tons. The change in cargo load will cause a significant change in the center of mass height, vehicle mass and front wheelbase, thus affecting the steering performance and rollover characteristics of the truck. In addition, the change in cargo load will also affect the vertical load of each wheel of the truck. There is a strong nonlinear relationship between the vertical load and the tire cornering stiffness. The change in vertical load affects the lateral force that the tire can provide, thereby changing the steering performance of the truck. The height of the center of mass of the sprung mass, the front wheelbase and the cornering stiffness of the front and rear wheels will change greatly due to the change in cargo load, and these parameters have a great influence on the steering performance and rollover characteristics of the vehicle. Therefore, accurately estimating the center of mass height, front wheelbase and cornering stiffness of the front and rear wheels under different cargo loads is of great significance to the safe driving of trucks.

[0003] The key to vehicle parameter estimation is to identify relevant parameters based on sensor signals and dynamic equations. Traditional parameter estimation methods mainly include: least squares method, polynomial method and interpolation method. However, when there is complex collinearity and imbalance between the parameters to be estimated, the estimation accuracy of the above traditional methods is low or even non-convergent. There is obvious complex collinearity and imbalance between the center of mass of the sprung mass of the truck, the front wheelbase and the lateral stiffness of the front and rear wheels, so the traditional methods cannot effectively estimate the above parameters. Summary of the invention

[0004] In view of the above-mentioned deficiencies of the prior art, the purpose of the present invention is to provide a method for truck parameter reconstruction under variable load based on RBF optimal ridge observer, so as to accurately estimate the height of the center of mass of the sprung mass, the front wheelbase and the lateral stiffness of the front and rear wheels of the current truck, thereby reconstructing a dynamic model with higher accuracy.

[0005] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0006] The method for reconstructing parameters of a truck under variable load based on an RBF optimal ridge observer of the present invention comprises the following steps:

[0007] 1) Based on the test data and the truck dynamics equation, the RBF neural network is trained to obtain the RBF optimal ridge parameter neural network that can calculate the optimal ridge parameter through the sensor signal;

[0008] 11) Establish the mathematical relationship between the parameters to be estimated and the measurable signals based on the truck's lateral dynamics equation and yaw dynamics equation;

[0009] 12) Collect and record the in-vehicle sensor data under different vehicle speeds, load capacities, and steering conditions through real vehicle tests, as well as the front and rear tire cornering stiffnesses under the corresponding conditions obtained through actual measurements; According to the collected data, calculate the optimal ridge parameter through the L-curve method; Save the sensor data under different conditions and the corresponding optimal ridge parameters as a data set;

[0010] 13) Train the RBF neural network based on the above data set. Use the sensor data as the input of the RBF neural network, and use the difference between the output value of the RBF neural network and the optimal ridge parameter as the loss function. Adopt the error backpropagation method to correct the hyperparameters of the RBF neural network, and obtain the RBF optimal ridge parameter neural network that can output the optimal ridge parameter based on the vehicle state through training;

[0011] 2) Estimate the front axle distance, sprung mass center height, and tire equivalent cornering stiffness based on the cross-iteration of basis vectors;

[0012] 21) Initialize the front axle distance, sprung mass center height, and tire equivalent cornering stiffness according to the parameter values in the half-loaded state of the truck;

[0013] 22) Judge whether the vehicle is currently in the starting stage according to the number of vehicle steering times; If it is the first steering condition, it is the starting stage and enter step 23) to estimate the front axle distance and sprung mass center height, and then enter step 24) to estimate the front and rear tire equivalent cornering stiffnesses; If it is not the first steering condition, the vehicle is in the non-starting stage and directly enter step 24);

[0014] 23) Measure the vehicle's yaw angle and center of mass sideslip angle in real time based on in-vehicle sensors; Estimate the front axle distance and sprung mass center height by using the forgetting least squares method based on the estimated tire equivalent cornering stiffness and the measured value of the center of mass sideslip angle;

[0015] 24) Take the basis vector composed of the front axle distance and sprung mass center height as a constant, calculate the optimal ridge parameter based on the RBF optimal ridge parameter neural network and substitute it into the ridge estimation algorithm to calculate the front and rear tire equivalent cornering stiffnesses;

[0016] Reconstruct the parameters of the truck dynamics model based on the estimated front axle distance, sprung mass center height, and tire equivalent cornering stiffness to make the dynamics model more in line with the current actual vehicle condition.

[0017] Furthermore, the mathematical relationship between the parameter to be estimated and the measurable signal in step 11) is derived from the truck lateral dynamics equation and the yaw dynamics equation, as shown in Equations (1) and (2):

[0018]

[0019]

[0020] where Y1 and Y2 are measurement matrices composed of measurement signals, Ф1 and Ф2 are model parameter matrices, X1 is a matrix composed of the front wheelbase and the height of the center of mass of the sprung mass, X2 is a matrix composed of the cornering stiffnesses of the front and rear wheels, M is the unsprung mass, m s is the sprung mass under no-load, △m s is the cargo mass, a is the front wheelbase, ω r is the yaw rate, L is the wheelbase, I z is the moment of inertia of the whole vehicle about the z-axis of the center of mass under no-load of the vehicle, I x is the moment of inertia of the unsprung mass about the x-axis under no-load, h s is the distance from the center of mass of the sprung mass with cargo to the roll axis, is the cornering stiffness of the front wheel, is the cornering stiffness of the rear wheel, β is the sideslip angle of the vehicle center of mass, δ is the front wheel steering angle, Δa is the increase in the front wheelbase caused by the change in the sprung mass, u is the vehicle speed, ω r is the yaw rate.

[0021] Furthermore, step 12) specifically includes: selecting different τ values at the current moment to obtain a set of points, approximately fitting this set of points to draw an L-shaped curve, and selecting the τ best value corresponding to the point with the maximum curvature on the curve as the ridge parameter; the ridge estimation parameter is as follows:

[0022] X2 = (Φ2 T Φ2 + τI) -1 Φ2 T Y2(3)

[0023] where I is the identity matrix and τ is the ridge parameter.

[0024] Furthermore, the in-vehicle sensor data in step 12) includes: yaw angular acceleration, yaw rate, sideslip angle of the center of mass, sideslip angular velocity of the center of mass, and vehicle speed; based on the above in-vehicle sensor data and the L-curve method, the mapping relationship between the current optimal ridge parameter τ best of the vehicle state is calculated.

[0025] Further, step 13) specifically includes: correcting the hyperparameters of the RBF neural network through the Adam-based backpropagation error method, continuously reducing the error to improve the estimation accuracy, and obtaining the mapping relationships of the accurate yaw rate, vehicle speed, mass, and optimal ridge parameter based on the following formulas (4)-(9);

[0026] J i = τ best - τ i (4)

[0027]

[0028]

[0029] m i = β1(m i-1 ) + (1 - β1)(ΔJ i )(7)

[0030]

[0031] v i = β2v i-1 + (1 - β2)(ΔJ i ) 2 (9)

[0032] where J i is the error value of the i-th iteration; τ i is the output value of the RBF neural network in the i-th iteration; S i is the hyperparameter in the RBF neural network; β1 and β2 are the exponential decay rates, which are respectively used to control the weight distribution and the influence of the gradient square, and are usually initialized to 0.9; ε is a constant; m i and v i respectively represent the exponentially weighted average of the past gradients and the exponentially weighted average of the squares of the gradients; m i and v i are the correction values, i is the number of iterations, and ΔJ is the error passed from the next layer network; during the training process, the model parameters are updated by formulas (4)-(9) in each iteration until the network error converges.

[0033] Further, in step 21), the half-loaded state of the truck means that the cargo load is half of the full load and the goods are evenly placed; the front wheelbase, the height of the center of mass of the sprung mass, the cornering stiffness of the front tire, and the cornering stiffness of the rear tire are measured in the stationary state; when the vehicle starts, the front wheelbase, the height of the center of mass of the sprung mass, the cornering stiffness of the front tire, and the cornering stiffness of the rear tire measured in the half-loaded state are used for parameter initialization.

[0034] Further, in step 22), the number of steering operations performed by the vehicle since startup is recorded by the steering wheel sensor. When the number of steering operations is greater than 1, it indicates that the front wheelbase and the height of the center of mass of the sprung mass have been estimated. Under the working conditions in the non-startup stage, only the real-time estimation of the front tire cornering stiffness and the rear tire cornering stiffness is performed.

[0035] Further, in step 23), based on the in-vehicle sensor data and the front and rear tire cornering stiffnesses estimated by the ridge estimation algorithm in the previous iteration, the matrix Φ1 and the measurement matrix Y1 are updated, and the expressions for estimating the front wheelbase and the height of the center of mass of the sprung mass using the forgetting least squares method are as follows:

[0036]

[0037] X1(t) = X1(t - 1) + K(t)(Y1(t) - Φ1(t)X1(t - 1))(11)

[0038]

[0039] where λ is the forgetting factor.

[0040] Further, in step 24), based on the optimal ridge parameter output by the basis vector and the RBF optimal ridge parameter neural network, the front and rear tire cornering stiffnesses are estimated by equations (2) and (3); through the cross-iteration method, all the above parameters to be estimated are gradually converged to the actual values to eliminate the influence of parameter imbalance on the estimation accuracy; the cross-iteration is specifically as follows: in step 23), the numerical values of the front and rear equivalent tire cornering stiffnesses are the results calculated in step 24) of the previous iteration, and the basis vector is updated by equations (10)-(12); in step 24), based on the basis vector, the RBF optimal ridge parameter, and the ridge estimation algorithm, the front and rear equivalent tire cornering stiffnesses are updated.

[0041] Advantages of the present invention:

[0042] Based on the existing in-vehicle sensors, the present invention estimates the front wheelbase, the height of the center of mass of the sprung mass, and the front and rear tire cornering stiffnesses in real time, thereby providing relatively accurate model parameters for model reconstruction and vehicle control, and improving the control efficiency and accuracy;

[0043] 1. Design an RBF optimal ridge parameter neural network to calculate the optimal ridge parameter most suitable for the current working conditions according to the data collected by the in-vehicle sensors in real time, thereby improving the dynamic accuracy of the estimation of the tire cornering stiffness and reducing the biased error caused by ridge estimation;

[0044] 2. Take the front wheelbase with time-invariant characteristics and the height of the center of mass of the sprung mass as base vectors, and estimate the front wheelbase, the height of the center of mass of the sprung mass, and the cornering stiffnesses of the front and rear tires through cross-iteration when the vehicle is in the starting stage, so as to eliminate the influence of the imbalance and multicollinearity among the above parameters on the estimation accuracy.

[0045] 3. When the vehicle is not in the starting stage, take the front wheelbase and the height of the center of mass of the sprung mass as constants, and only estimate the cornering stiffnesses of the front and rear tires in real time, thereby improving the estimation efficiency. Brief Description of the Drawings

[0046] Figure 1 It is the flowchart of the RBF optimal ridge parameter neural network training in the present invention.

[0047] Figure 2 It is the flowchart of the parameter estimation algorithm based on the cross-iteration of base vectors in the present invention. Detailed Embodiment

[0048] For the convenience of understanding by those skilled in the art, the present invention will be further described below in conjunction with the embodiments and the drawings. The content mentioned in the embodiments does not limit the present invention.

[0049] A method for parameter reconstruction under variable loads of a truck based on an RBF optimal ridge observer of the present invention comprises the following steps:

[0050] 1) Train an RBF neural network based on test data and the truck dynamics equation to obtain an RBF optimal ridge parameter neural network capable of calculating the optimal ridge parameter through sensor signals; refer to Figure 1 as shown;

[0051] 11) Establish the mathematical relationship between the parameters to be estimated and the measurable signals based on the truck lateral dynamics equation and the yaw dynamics equation;

[0052] The mathematical relationship between the parameters to be estimated and the measurable signals is derived from the truck lateral dynamics equation and the yaw dynamics equation, as shown in Equations (1) and (2):

[0053]

[0054]

[0055] where Y1 and Y2 are measurement matrices composed of measurement signals, Ф1 and Ф2 are model parameter matrices, X1 is a matrix composed of the front wheelbase and the height of the center of mass of the sprung mass, X2 is a matrix composed of the cornering stiffnesses of the front and rear wheels, M is the unsprung mass, m s is the sprung mass under no-load, △m s is the cargo mass, a is the front wheelbase, ω ris the yaw rate, L is the wheelbase, and I z is the moment of inertia of the entire vehicle about the z-axis of the center of mass when the vehicle is unloaded, and I x is the moment of inertia of the unsprung mass about the x-axis when unloaded, and h s is the distance from the center of mass of the sprung mass with load to the roll axis, is the cornering stiffness of the front wheels, is the cornering stiffness of the rear wheels, β is the sideslip angle of the vehicle center of mass, δ is the front wheel steering angle, Δa is the increase in the front wheelbase caused by the change in the sprung mass, u is the vehicle speed, and ω r is the yaw rate.

[0056] 12) Collect and record the in-vehicle sensor data under different vehicle speeds, load weights, and steering conditions through real vehicle tests, as well as the cornering stiffness of the front wheels and the cornering stiffness of the rear wheels under the corresponding conditions obtained through actual measurements; according to the collected data, calculate the optimal ridge parameter through the L-curve method; save the sensor data under different conditions and the corresponding optimal ridge parameter as a data set;

[0057] The in-vehicle sensor data includes: yaw angular acceleration, yaw rate, sideslip angle of the center of mass, sideslip angular velocity of the center of mass, and vehicle speed; calculate the current optimal ridge parameter τ of the vehicle state based on the above in-vehicle sensor data and the L-curve method best of the mapping relationship.

[0058] Select different τ values at the current moment to obtain a set of points, approximately fit this set of points to draw an L-shaped curve, and select the τ best value corresponding to the point with the maximum curvature in the curve as the ridge parameter; the ridge estimation parameter is as follows:

[0059] X2 = (Φ2 T Φ2 + τI) -1 Φ2 T Y2(3)

[0060] where I is the identity matrix and τ is the ridge parameter.

[0061] 13) Train the RBF neural network based on the above data set, use the sensor data as the input of the RBF neural network, use the difference between the output value of the RBF neural network and the optimal ridge parameter as the loss function, and use the error backpropagation method to correct the hyperparameters of the RBF neural network, and obtain the RBF optimal ridge parameter neural network that can output the optimal ridge parameter based on the vehicle state through training;

[0062] Correct the hyperparameters of the RBF neural network through the backpropagation of error based on Adam, continuously reduce the error and improve the estimation accuracy, and obtain the accurate mapping relationship of yaw rate, vehicle speed, mass, and optimal ridge parameter based on the following formulas (4)-(9);

[0063] J i = τ best -τ i (4)

[0064]

[0065]

[0066] m i = β1(m i-1 ) + (1 - β1)(ΔJ i )(7)

[0067]

[0068] v i = β2v i-1 + (1 - β2)(ΔJ i ) 2 (9)

[0069] Wherein, J i is the error value of the i-th iteration; τ i is the output value of the RBF neural network in the i-th iteration; S i is the hyperparameter in the RBF neural network; β1 and β2 are the exponential decay rates, which are respectively used to control the weight distribution and the influence of the gradient square, and are usually initialized to 0.9; ε is a constant; and respectively represent the exponentially weighted average of the past gradient and the exponentially weighted average of the square of the gradient; m i and v i are correction values, i is the number of iterations, and ΔJ is the error passed from the next layer network; during the training process, the model parameters are updated by equations (4)-(9) in each iteration until the network error converges.

[0070] 2) Estimate the front axle distance, the height of the center of mass of the sprung mass, and the equivalent cornering stiffness of the tire based on the cross-iteration of the basis vectors; as shown in reference Figure 2 ;

[0071] 21) Initialize the front axle distance, the height of the center of mass of the sprung mass, and the equivalent cornering stiffness of the tire according to the parameter values in the half-loaded state of the truck;

[0072] The half-loaded state of the truck means that the cargo load is half of the full load and the goods are evenly placed; measure the front axle distance, the height of the center of mass of the sprung mass, the cornering stiffness of the front tire, and the cornering stiffness of the rear tire in the stationary state; when the vehicle starts, initialize the parameters of the front axle distance, the height of the center of mass of the sprung mass, the cornering stiffness of the front tire, and the cornering stiffness of the rear tire measured in the above-mentioned half-loaded state.

[0073] 22) Determine whether the vehicle is currently in the starting stage based on the number of vehicle steering operations; if it is the first steering condition, it is in the starting stage and proceed to step 23) to estimate the front wheelbase and the height of the center of mass of the sprung mass, and then proceed to step 24) to estimate the equivalent cornering stiffness of the front and rear tires; if it is not the first steering condition, the vehicle is in the non-starting stage and directly proceed to step 24);

[0074] Among them, the number of steering operations performed by the vehicle since startup is recorded by the steering wheel sensor. When the number of steering operations is greater than 1, it indicates that the front wheelbase and the height of the center of mass of the sprung mass have been estimated. Since the front wheelbase and the height of the center of mass of the sprung mass are time-invariant during vehicle driving, under the non-starting stage conditions, only the cornering stiffness of the front tire and the cornering stiffness of the rear tire are estimated in real time.

[0075] 23) Based on the on-vehicle sensors, the yaw angle and the sideslip angle of the center of mass of the whole vehicle are measured in real time; based on the estimated equivalent cornering stiffness of the tires and the measured value of the sideslip angle of the center of mass, the forgetting least squares method is used to estimate the front wheelbase and the height of the center of mass of the sprung mass;

[0076] Based on the on-vehicle sensor data and the cornering stiffness update matrix Φ1 and the measurement matrix Y1 of the front and rear tires estimated by the ridge estimation algorithm in the previous iteration, the expressions for estimating the front wheelbase and the height of the center of mass of the sprung mass using the forgetting least squares method are as follows:

[0077]

[0078] X1(t) = X1(t - 1) + K(t)(Y1(t) - Φ1(t)X1(t - 1)) (11)

[0079]

[0080] Among them, λ is the forgetting factor.

[0081] 24) Take the basis vectors composed of the front wheelbase and the height of the center of mass of the sprung mass as constants, calculate the optimal ridge parameter based on the RBF optimal ridge parameter neural network and substitute it into the ridge estimation algorithm to calculate the equivalent cornering stiffness of the front tire and the equivalent cornering stiffness of the rear tire;

[0082] Based on the optimal ridge parameter of the neural network output with basis vectors and the optimal RBF ridge parameter, the front and rear cornering stiffnesses are estimated through Equation (2) and Equation (2); all the parameters to be estimated are gradually converged to the actual values by means of cross-iteration, eliminating the influence of parameter imbalance on the estimation accuracy; the specific cross-iteration is as follows: in step 23), the numerical values of the front and rear equivalent cornering stiffnesses are the results calculated in step 24) of the previous iteration, and the basis vectors are updated through Equations (10)-(12); in step 24), based on the basis vectors, the optimal RBF ridge parameter and the ridge estimation algorithm, the front and rear equivalent cornering stiffnesses are updated.

[0083] Based on the estimated front wheelbase, the height of the center of mass of the sprung mass, and the tire equivalent cornering stiffness, the parameters of the truck dynamics model are reconstructed to make the dynamics model more in line with the current actual vehicle condition.

[0084] There are many specific application ways of the present invention. The above description is only the preferred embodiment of the present invention. It should be pointed out that for those of ordinary skill in the art of this technology, without departing from the principle of the present invention, several improvements can be made, and these improvements should also be regarded as the protection scope of the present invention.

Claims

1. A parameter reconstruction method for a truck under variable load based on an RBF optimal ridge observer, characterized in that, The steps are as follows: 1) Train the RBF neural network based on the test data and the truck dynamics equation to obtain the RBF optimal ridge parameter neural network that can calculate the optimal ridge parameter through the sensor signal; 11) Establish the mathematical relationship between the parameter to be estimated and the measurable signal based on the truck lateral dynamics equation and the yaw dynamics equation; 12) Collect and record the on-vehicle sensor data under different vehicle speeds, load capacities, and steering conditions through real vehicle tests, and the front and rear tire cornering stiffnesses under the corresponding conditions obtained through actual measurement; calculate the optimal ridge parameter by the L-curve method according to the collected data; Save the sensor data under different conditions and the corresponding optimal ridge parameters as a data set; 13) Train the RBF neural network based on the above data set, use the sensor data as the input of the RBF neural network, use the difference between the output value of the RBF neural network and the optimal ridge parameter as the loss function, and use the error backpropagation method to correct the hyperparameters of the RBF neural network, and obtain the RBF optimal ridge parameter neural network that can output the optimal ridge parameter based on the vehicle state through training; 2) Estimate the front axle distance, the height of the center of mass of the sprung mass, and the equivalent tire cornering stiffness based on the cross-iteration of the basis vectors; 21) Initialize the front axle distance, the height of the center of mass of the sprung mass, and the equivalent tire cornering stiffness according to the parameter values in the half-loaded state of the truck; 22) Judge whether the vehicle is in the starting stage according to the number of vehicle steering; if it is the first steering condition, it is the starting stage and enter step 23) to estimate the front axle distance and the height of the center of mass of the sprung mass, and then enter step 24) to estimate the equivalent cornering stiffness of the front and rear tires; if it is not the first steering condition, the vehicle is in the non-starting stage and directly enter step 24); 23) Measure the vehicle yaw angle and the center of mass sideslip angle in real time based on the on-vehicle sensors; estimate the front axle distance and the height of the center of mass of the sprung mass by the forgetting least squares method based on the estimated equivalent tire cornering stiffness and the measured value of the center of mass sideslip angle; 24) Take the basis vector composed of the front axle distance and the height of the center of mass of the sprung mass as a constant, calculate the optimal ridge parameter based on the RBF optimal ridge parameter neural network and substitute it into the ridge estimation algorithm to calculate the equivalent cornering stiffness of the front tire and the rear tire; Reconstruct the parameters of the truck dynamics model based on the estimated front axle distance, the height of the center of mass of the sprung mass, and the equivalent tire cornering stiffness, so that the dynamics model better conforms to the current actual vehicle conditions.

2. The parameter reconstruction method for a truck under variable load based on an RBF optimal ridge observer according to claim 1, characterized in that, The mathematical relationship between the parameter to be estimated and the measurable signal in step 11) is derived from the truck lateral dynamics equation and the yaw dynamics equation, as shown in equations (1) and (2): Among them, Y1 and Y2 are measurement matrices composed of measurement signals, Φ1 and Φ2 are model parameter matrices, X1 is a matrix composed of the front axle distance and the height of the center of mass of the sprung mass, X2 is a matrix composed of the cornering stiffnesses of the front and rear wheels, M is the unsprung mass, m s is the sprung mass under no-load, △m s is the cargo mass, a is the front axle distance, ω r is the yaw rate, L is the wheelbase, I z is the moment of inertia of the whole vehicle about the z-axis of the center of mass under no-load of the vehicle, I x is the moment of inertia of the unsprung mass about the x-axis under no-load, h s is the distance from the center of mass of the sprung mass with cargo to the roll axis, is the cornering stiffness of the front wheel, is the cornering stiffness of the rear wheel, β is the sideslip angle of the vehicle center of mass, δ is the front wheel steering angle, Δa is the increase in the front axle distance caused by the change in the sprung mass, u is the vehicle speed, ω r is the yaw rate.

3. The parameter reconstruction method for a truck under variable load based on an RBF optimal ridge observer according to claim 2, characterized in that, Specifically included in the said step 12) is: selecting different τ values at the current moment to obtain a set of points, approximately fitting this set of points to draw an L-shaped curve, and selecting the τ best value corresponding to the point with the maximum curvature in the curve as the ridge parameter; the ridge estimation parameter is as follows: X2 = (Φ2 T Φ2 + τI) -1 Φ2 T Y2 (3) Where, I is the identity matrix, and τ is the ridge parameter.

4. The parameter reconstruction method for a truck under variable load based on an RBF optimal ridge observer according to claim 3, characterized in that, The in-vehicle sensor data in step 12) includes: yaw angular acceleration, yaw angular velocity, sideslip angle of the center of mass, sideslip angular velocity of the center of mass, and vehicle speed; based on the above in-vehicle sensor data and the L-curve method, the current optimal ridge parameter τ of the vehicle state is calculated best The mapping relationship between 5. The parameter reconstruction method for a truck under variable load based on an RBF optimal ridge observer according to claim 3, characterized in that, Step 13) specifically includes: correcting the hyperparameters of the RBF neural network by the backpropagation of error based on Adam, continuously reducing the error to improve the estimation accuracy, and training to obtain an accurate mapping relationship between the yaw angular velocity, vehicle speed, mass, and optimal ridge parameter based on the following equations (4)-(9); J i = τ best -τ i (4) m i = β1(m i-1 ) + (1 - β1)(ΔJ i ) (7) v i = β2v i-1 + (1 - β2)(ΔJ i ) 2 (9) Among them, J i is the error value of the i-th iteration; τ i is the output value of the RBF neural network in the i-th iteration; S i is the hyperparameter in the RBF neural network; β1 and β2 are the exponential decay rates, which are used to control the influence of weight allocation and the square of the gradient respectively, and are usually initialized to 0.9; ε is a constant; and respectively represent the exponentially weighted average of the past gradients and the exponentially weighted average of the square of the gradient; m i and v i are correction values, i is the number of iterations, and ΔJ is the error passed from the next layer network; during the training process, the model parameters are updated by equations (4)-(9) in each iteration until the network error converges.

6. The parameter reconstruction method for a truck under variable load based on an RBF optimal ridge observer according to claim 3, characterized in that, In step 23), based on the in-vehicle sensor data and the front and rear tire cornering stiffnesses estimated by the ridge estimation algorithm in the previous iteration, update the matrix Φ1 and the measurement matrix Y1, and use the forgetting least squares method to estimate the front wheelbase and the height of the center of mass of the sprung mass. The expressions are as follows: X1(t) = X1(t - 1) + K(t)(Y1(t) - Φ1(t)X1(t - 1)) (11) where λ is the forgetting factor.

7. The parameter reconstruction method under variable load of a truck based on the RBF optimal ridge observer according to claim 6, characterized in that, In step 24), based on the optimal ridge parameter output by the neural network of the basis vector and the RBF optimal ridge parameter, estimate the front and rear tire cornering stiffnesses through equations (2) and (3); make all the above parameters to be estimated gradually converge to the actual values through the cross-iteration method to eliminate the influence of parameter imbalance on the estimation accuracy; the specific cross-iteration is as follows: in step 23), the values of the equivalent cornering stiffnesses of the front and rear tires are the results calculated in step 24) of the previous iteration, and update the basis vector through equations (10)-(12); in step 24), based on the basis vector, the RBF optimal ridge parameter and the ridge estimation algorithm, update the equivalent cornering stiffnesses of the front and rear wheels.

8. The parameter reconstruction method under variable load of a truck based on the RBF optimal ridge observer according to claim 1, characterized in that, In step 21), the half-loaded state of the truck means that the cargo load is half of the full load and the cargo is evenly placed; measure the front wheelbase, the height of the center of mass of the sprung mass, the front tire cornering stiffness, and the rear tire cornering stiffness under the stationary state; when the vehicle starts, initialize the parameters of the front wheelbase, the height of the center of mass of the sprung mass, the front tire cornering stiffness, and the rear tire cornering stiffness measured under the half-loaded state.

9. The parameter reconstruction method under variable load of a truck based on the RBF optimal ridge observer according to claim 1, characterized in that, In step 22), record the number of steering operations of the vehicle since startup through the steering wheel sensor. When the number of steering operations is greater than 1, it indicates that the front wheelbase and the height of the center of mass of the sprung mass have been estimated. Under the working conditions in the non-starting stage, only the front tire cornering stiffness and the rear tire cornering stiffness are estimated in real time.

Images