Adaptive identification method of dynamic parameters of autonomous driving vehicle based on SQP and GRNN

Abstract

The present invention belongs to the field of vehicle dynamics technology and relates to a method for adaptively identifying the dynamic parameters of an autonomous vehicle based on SQP and GRNN. The method comprises four parts: establishing a mechanism model of the autonomous vehicle, establishing a GRNN model, obtaining a training set based on the SQP algorithm, and adaptively identifying parameters based on the trained GRNN model. The mechanism model of the autonomous vehicle and the SQP algorithm are used to solve the key parameters of the dynamic model that are difficult to obtain using conventional methods offline, and the solution results are used as the training set for the GRNN model. This enables the trained GRNN model to have the ability to handle the nonlinear relationship between the steering wheel angle, vehicle speed, and tire lateral stiffness, suspension roll stiffness, and roll damping in the mechanism model of the autonomous vehicle. Ultimately, accurate dynamic parameters are identified in real time based on the vehicle's operating conditions, laying the foundation for the effective control of the autonomous vehicle.

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicle dynamics and relates to a method for adaptively identifying dynamic parameters of an autonomous driving vehicle based on SQP and GRNN. Background Art

[0002] In recent years, the development of autonomous vehicles has become a prominent topic in the field of intelligent transportation. Compared to traditional human-driven vehicles, autonomous vehicles offer significant advantages in terms of improved safety, comfort, and economy. When operating autonomous vehicles, not only must the vehicle's trajectory tracking capability be considered, but also its driving stability. This is particularly true for vehicles like buses and SUVs, which carry large numbers of passengers and have high centers of mass. Accidents can lead to even greater economic losses and casualties. Effective control, the primary approach to achieving safe autonomous vehicle operation, relies on a mechanistic model that reflects the vehicle's actual dynamic characteristics and accurate dynamic parameters. While parameters in the mechanistic model, such as vehicle speed and steering wheel angle, can be measured using sensors, parameters such as tire cornering stiffness, suspension roll stiffness, and roll damping cannot be acquired using conventional methods. Therefore, accurately acquiring these parameters is crucial to improving the safety of autonomous vehicles.

[0003] To obtain these key parameters, experts and scholars have combined optimization algorithms and economical sensors to identify or estimate dynamic parameters that are difficult to measure directly. The Chinese patent application number 202310958288.2 proposes a tire cornering stiffness estimation method, which reasonably combines a variety of on-board sensors to obtain parameters such as the vehicle's longitudinal speed, lateral acceleration, and yaw angular velocity, and establishes a tire model based on mathematical derivation to estimate the tire's cornering stiffness based on vehicle driving information. However, this patent mainly estimates the tire's cornering stiffness and does not take into account parameters such as the suspension roll stiffness and roll damping in the complex mechanism model.

[0004] In addition, a paper titled "Research on Parameter Identification of Simplified Heavy-Duty Semi-trailer Models," published in the journal Automotive Engineering, Vol. 37, No. 6, 2015, established three-degree-of-freedom and five-degree-of-freedom dynamic models for heavy-duty vehicles. The paper thoroughly analyzed the impact of key model parameters on the output results and used a genetic algorithm to offline identify key parameters such as tire cornering stiffness, suspension roll stiffness, and roll damping at different steering wheel angles and vehicle speeds. The identified data was then fitted using linear interpolation, enabling online determination of key model parameters based on the vehicle's real-time state. This approach laid the foundation for parameter estimation and vehicle stability control. However, this method requires storing a large amount of known data in the controller, which imposes a significant storage burden. Furthermore, the complex nonlinear relationship between the vehicle's steering wheel angle, speed, and dynamic parameters compromises the accuracy of parameter identification using linear interpolation. Summary of the Invention

[0005] The purpose of the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a method for adaptively identifying the dynamic parameters of an autonomous vehicle based on a Sequential Quadratic Programming Algorithm (SQP) and a Generalized Regression Neural Network (GRNN), which lays the foundation for obtaining a high-precision dynamic model and effectively controlling the autonomous vehicle.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for adaptively identifying the dynamic parameters of an autonomous vehicle based on SQP and GRNN includes four parts: establishing a mechanism model of the autonomous vehicle, establishing a GRNN model, obtaining a training set based on an SQP algorithm, and adaptively identifying parameters based on the trained GRNN model. The accuracy of the adaptive identification of the trained GRNN model is verified by an adaptive parameter hybrid dynamic model, which consists of two parts: the trained GRNN model and the mechanism model of the autonomous vehicle.

[0008] The establishment of the autonomous driving vehicle mechanism model is specifically as follows: establishing a three-degree-of-freedom model of the autonomous driving vehicle including the vehicle's lateral motion, yaw motion, and roll motion, and its differential equation expression is specifically as follows:

[0009]

[0010] In formula (1): v y is the lateral velocity; k f is the front axle tire cornering stiffness; k r is the rear axle tire cornering stiffness; I x is the moment of inertia of the sprung mass around the x-axis; I xs =I x +m s h 2 is the moment of inertia of the suspension sprung mass around the roll axis; m is the vehicle mass; m s is the sprung mass; h is the height from the center of mass to the roll center; δ f is the front wheel angle, which is linearly related to the steering wheel angle δ; v x is the longitudinal velocity; a is the distance from the center of mass to the front axle; b is the distance from the center of mass to the rear axle; ω is the yaw angular velocity; I z is the moment of inertia of the sprung mass around the z-axis; K s is the suspension roll stiffness; C s is the suspension roll damping; φ is the roll angle; g is the acceleration due to gravity.

[0011] Due to the steering wheel angle δ and the longitudinal velocity v x They can be easily obtained by the steering angle sensor and vehicle speed sensor respectively, so the steering wheel angle δ and longitudinal speed v x As input to the mechanism model.

[0012] The GRNN model is established by constructing a GRNN model including an input layer, a pattern layer, a summation layer, and an output layer. The input layer parameters include variables that can reflect the vehicle's driving state, such as the steering wheel angle δ, the longitudinal speed v x etc.; the output layer parameters include key dynamic parameters in the autonomous vehicle mechanism model, such as the front axle tire lateral stiffness k f , rear axle tire cornering stiffness k r and suspension roll stiffness K s , roll damping C s wait.

[0013] The training set acquisition based on the SQP algorithm is specifically as follows: using the SQP optimization algorithm, the front axle tire cornering stiffness k corresponding to the autonomous driving vehicle mechanism model under typical steering wheel angles and vehicle speeds is obtained. f , rear axle tire cornering stiffness k r and suspension roll stiffness K s , suspension roll damping C s The parameter values ​​of are used as the training data set for the GRNN model.

[0014] Furthermore, the SQP algorithm solves the problem by finding the key parameter values ​​in the autonomous vehicle's mechanics model so that the center of mass slip angle β, yaw rate ω, and roll angle φ output by the autonomous vehicle's mechanics model are as close as possible to the corresponding dynamic indicators of the actual vehicle. The specific optimization problem can be described as follows:

[0015]

[0016] In formula (2), is the i-th center of mass sideslip angle in the time series output by the autonomous vehicle mechanism model; is the sideslip angle of the i-th center of mass in the actual vehicle driving time series; is the i-th yaw rate in the time series output by the autonomous vehicle mechanism model; is the ith yaw rate in the actual vehicle driving time series; is the i-th roll angle in the time series output by the autonomous vehicle mechanism model; is the i-th roll angle in the actual vehicle driving time series; α, γ and ξ are weight coefficients respectively.

[0017] Furthermore, the solution conditions obtained from the training set of the SQP algorithm are selected from the typical operating conditions of autonomous vehicles, and the steering wheel angle and longitudinal speed range under these conditions should cover the conditions of most autonomous vehicles when driving normally. In order to make the solution results serve as the training set of the GRNN model, the solution process is divided into n time series. The longitudinal speed corresponding to the i-th time series under the solution condition can be expressed as v xi , the corresponding steering wheel angle is expressed as δ i The corresponding front axle tire cornering stiffness, rear axle tire cornering stiffness, suspension roll stiffness and suspension roll damping are expressed as k fi , k ri , K si , C si After solving the problem, we finally obtained n sets of data that can represent the relationship between steering wheel angle, longitudinal speed and tire cornering stiffness (front axle tire cornering stiffness, rear axle tire cornering stiffness), suspension roll stiffness and suspension roll damping, and used them as the training set of the GRNN model.

[0018] The parameter adaptive identification based on the trained GRNN model is specifically as follows: the operating conditions of the autonomous vehicle and the corresponding SQP solution results are used as the training set of the GRNN model to learn the steering wheel angle δ and longitudinal speed v of the autonomous vehicle. x The nonlinear relationship between the key parameters of the autonomous vehicle mechanism model is that the trained GRNN model can predict the vehicle's steering wheel angle δ and longitudinal speed v x Online identification of the front axle tire cornering stiffness k f , rear axle tire cornering stiffness k r and suspension roll stiffness K s and roll damping C s , to achieve adaptive identification of dynamic parameters of autonomous vehicles.

[0019] Furthermore, the principle of parameter adaptive identification based on the trained GRNN model is as follows: the SQP solution result is used as the training data set of the GRNN model, and the input data is set as the steering wheel angle δ and the longitudinal speed v x , the output data is the front axle tire cornering stiffness k f , rear axle tire cornering stiffness k r and suspension roll stiffness K s and roll damping C s; Then, the input layer is constructed to determine the dimension of the input data, and the input data is directly passed to the pattern layer. The pattern layer clusters the input data based on its dimension, uses the Gaussian function to calculate the weight between the input data and each training set sample, multiplies the output data in each training set sample by its corresponding weight, and then the summation layer adds all the results, and finally the output layer outputs the weighted average of all the results. After the above training process, the GRNN model can be used to calculate the steering wheel angle δ and longitudinal speed v according to the input data. x , real-time identification and output of the front axle tire cornering stiffness k f , rear axle tire cornering stiffness k r and suspension roll stiffness K s and suspension roll damping C s .

[0020] Compared with the prior art, the technical solution provided by the present invention has the following beneficial effects:

[0021] This adaptive identification method of autonomous vehicle dynamic parameters based on SQP and GRNN can use the autonomous vehicle mechanism model and SQP algorithm to solve the key parameters of the dynamic model that are difficult to obtain by conventional methods offline; the SQP solution results are used as the training set of the GRNN model, so that the trained GRNN model has the ability to process the vehicle steering wheel angle δ, longitudinal speed v x and tire cornering stiffness (k f 、k r ), suspension roll stiffness K s and suspension roll damping C s Finally, the trained GRNN model can identify accurate dynamic parameters (front axle tire cornering stiffness k) in real time according to the vehicle's operating conditions. f , rear axle tire cornering stiffness k r and suspension roll stiffness K s and suspension roll damping C s ), laying the foundation for the effective control of self-driving cars. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings are incorporated in and constitute a part of this specification and, together with the description, serve to explain the principles of the invention.

[0023] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0024] Figure 1This is a block diagram of the principle of the adaptive identification method of dynamic parameters of an autonomous driving vehicle based on SQP and GRNN provided by the present invention;

[0025] Figure 2 A schematic diagram of the three-degree-of-freedom mechanism model of the autonomous driving vehicle provided by the present invention;

[0026] Figure 3 A schematic diagram of the adaptive parameter hybrid dynamics model for an autonomous vehicle provided by the present invention;

[0027] Figure 4 This is the curve of the sideslip angle of the center of mass of the autonomous vehicle changing with time in the simulation test;

[0028] Figure 5 This is the curve of the yaw rate of the autonomous bus changing with time in the simulation test;

[0029] Figure 6 This is the curve of the roll angle changing with time in the simulation test of the autonomous driving car. DETAILED DESCRIPTION

[0030] Here, exemplary embodiments will be described in detail, and the embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are only examples consistent with some aspects of the present invention described in detail in the appended claims.

[0031] In order to enable those skilled in the art to better understand the technical solutions of the present invention, the present invention is described in further detail below with reference to the accompanying drawings and embodiments.

[0032] Example

[0033] The principle of the self-driving car adaptive parameter identification method based on SQP and GRNN provided in this embodiment is as follows: Figure 1 As shown:

[0034] First, a three-degree-of-freedom mechanism model is established that can accurately describe the posture changes of the autonomous vehicle, such as Figure 2 As shown, there are three degrees of freedom: lateral, yaw, and roll. To facilitate the establishment of the model, the following assumptions are made:

[0035] (1) Ignore the influence of the steering system and use the front wheel angle as input;

[0036] (2) The motion of the vehicle body in the pitch direction is not considered;

[0037] (3) Ignore the impact of aerodynamic factors on vehicle status;

[0038] (4) Assume that the road on which the autonomous vehicle is traveling is a horizontal road and ignore the vertical movement of the vehicle;

[0039] (5) Ignore the changes in tire characteristics and aligning torque caused by changes in left and right tire loads.

[0040] The moment balance equation along the y-axis is:

[0041]

[0042] The moment balance equation about the z-axis is:

[0043]

[0044] The moment balance equation about the x-axis is:

[0045]

[0046] In formula (3) to formula (5), v y is the lateral velocity; F y1 is the lateral force on the front axle tire, F y2 is the lateral force on the rear axle tire; a y is the lateral acceleration; I x is the moment of inertia of the sprung mass around the x-axis; I xs =I x +m s h 2 is the moment of inertia of the suspension sprung mass around the roll axis; m is the vehicle mass; m s is the sprung mass; h is the height from the center of mass to the roll center; δ f is the front wheel turning angle; v x is the longitudinal velocity (i.e., vehicle speed); a is the distance from the center of mass to the front axle; b is the distance from the center of mass to the rear axle; ω is the yaw angular velocity; I z is the moment of inertia of the sprung mass around the z-axis; K s is the suspension roll stiffness; C s is the suspension roll damping; φ is the roll angle; g is the acceleration due to gravity.

[0047] The relationship between tire lateral force and sideslip angle is:

[0048]

[0049] In formula (6), α f is the front wheel slip angle; α r is the rear wheel slip angle; k f is the front axle tire cornering stiffness; k r is the rear axle tire cornering stiffness.

[0050] Under the assumption of a small tire slip angle, the following kinematic equations hold:

[0051]

[0052] In formula (7), β≈v y / v x is the vehicle's sideslip angle.

[0053] After sorting, the differential equation expression of the autonomous driving vehicle mechanism model is obtained, as shown in formula (1).

[0054] Furthermore, a GRNN model is established, which includes four layers of neurons, namely input layer, pattern layer, summation layer and output layer. The corresponding network input is X = [x1, x2, ..., x n ] T , the output is Y=[y1,y2,…,y k ] T The following describes the four-layer structure of the GRNN model:

[0055] (1) Input layer

[0056] The number of neurons in the input layer is equal to the dimension of the input data in the learning sample. Each neuron is a simple distribution unit that directly passes the variable to the pattern layer. x and steering wheel angle δ are used as input data for the GRNN model, namely

[0057]

[0058] (2) Model layer

[0059] The number of neurons in the pattern layer is equal to the number of training samples n. Each neuron corresponds to a different sample. The transfer function of the neuron in the pattern layer is:

[0060]

[0061] In formula (9), X is the network input data; X i is the learning sample corresponding to the i-th neuron.

[0062] (3) Summation layer: Two types of neurons are used for summation in the summation layer.

[0063] The calculation formula for the first category is:

[0064]

[0065] Formula (10) performs arithmetic summation on the outputs of all pattern layer neurons. The connection weight between the pattern layer and each neuron is 1, and the transfer function S D Expressed as:

[0066]

[0067] The calculation formula for the second category is:

[0068]

[0069] Formula (12) performs weighted summation on the neurons of all pattern layers, and the connection weight between the i-th neuron in the pattern layer and the j-th molecular summation neuron in the summation layer is y ij , transfer function S Nj Expressed as:

[0070]

[0071] In formula (13), y ij is the jth element in the i-th output data.

[0072] (4) Output layer

[0073] The number of neurons in the output layer is equal to the dimension k of the output data in the learning sample. Each neuron divides the output of the summation layer. The output of neuron j corresponds to the estimated result The jth element of

[0074]

[0075] Furthermore, the SQP optimization algorithm is used to optimize the front axle tire cornering stiffness k in the autonomous vehicle mechanism model. f , rear axle tire cornering stiffness k r and suspension roll stiffness K s and suspension roll damping C s Perform offline solution and use the solution result as the training data set of GRNN model. The general optimization problem of SQP optimization algorithm can be expressed as:

[0076]

[0077] In formula (15), f(x) is the optimization objective function, h i (x) = 0 represents an equality constraint; g i (x)≥0 represents an inequality constraint.

[0078] Furthermore, the above-mentioned process of solving the parameters of the autonomous vehicle mechanism model based on the SQP algorithm is to obtain the corresponding sideways stiffness, roll stiffness, and roll damping values, so that the output of the autonomous vehicle mechanism model is as close as possible to the dynamic indicators of the actual vehicle. Considering the high cost of using real vehicle experiments and the greater danger during high-speed driving, this embodiment uses the vehicle model in the professional software TruckSim to replace the actual vehicle and output the corresponding dynamic indicators. In the problem of autonomous vehicle motion control, the three commonly used indicators that can characterize the vehicle's dynamic characteristics are the sideslip angle β, the yaw angular velocity ω, and the roll angle φ. Therefore, the solution process can be transformed into the following nonlinear optimization problem:

[0079]

[0080] In formula (16), is the i-th center of mass sideslip angle in the time series output by the autonomous vehicle mechanism model; The i-th center of mass sideslip angle in the time series output by the TruckSim model; is the i-th yaw rate in the time series output by the autonomous vehicle mechanism model; is the i-th yaw rate in the time series output by the TrcukSim model; is the i-th roll angle in the time series output by the autonomous vehicle mechanism model; is the i-th roll angle in the time series output by the TruckSim model; α, γ, and ξ are weight coefficients, which in this embodiment are α=4, γ=3, and ξ=1, respectively, and are determined based on experience.

[0081] Furthermore, in this embodiment, the solution condition of the SQP optimization algorithm is selected as the angle step condition. When the vehicle speed is lower than 10km / h, the autonomous vehicle generally does not become unstable, and under low-speed conditions, the tire and suspension characteristics are in the linear region, so the solution speed is selected from 20km / h to 60km / h, with an interval of 10km / h. This speed range covers the driving requirements of most autonomous vehicles and can meet actual needs. The steering wheel angle is selected from -150° to 150°, with an interval of 10°. After solving, the lateral stiffness, roll stiffness and roll damping under a series of working conditions are finally obtained. The solution process is divided into n time series, and the longitudinal velocity under the solution condition corresponding to the i-th time series can be expressed as v xi , the corresponding steering wheel angle is expressed as δ i The corresponding front axle tire cornering stiffness, rear axle tire cornering stiffness, suspension roll stiffness and roll damping are expressed as k fi , k ri , K si , Csi (i=1,2,…,n). After the above solution process, a total of 154 sets of data are obtained as the GRNN model training set.

[0082] Furthermore, in order to verify the accuracy of the adaptive identification of the trained GRNN model, the following Figure 3 The self-driving car adaptive parameter hybrid dynamics model shown in the figure consists of two parts: the trained GRNN model and the self-driving car mechanism model. The input of the self-driving car adaptive parameter hybrid dynamics model is the longitudinal speed v of the self-driving car. x and steering wheel angle δ. Among them, the longitudinal velocity v x and steering wheel angle δ are derived from the TruckSim model. Passing these two sets of parameters (the aforementioned input data) into the trained GRNN model allows real-time identification of the tire cornering stiffness, suspension roll stiffness, and suspension roll damping under the vehicle's current operating conditions. These parameters serve as inputs to the autonomous vehicle's mechanics model, which then calculates the vehicle's slip angle β, yaw rate ω, and roll angle φ in real time.

[0083] In order to verify the effectiveness of the proposed adaptive identification method for the dynamic parameters of an autonomous vehicle based on SQP and GRNN, a relevant model was built and an algorithm was developed in the TruckSim and Matlab / Simulink joint simulation platform. Simulation tests were conducted under double lane-changing conditions, and the test results were compared with the output results of the TruckSim model and the fixed-parameter dynamic model.

[0084] Some vehicle parameters used in this embodiment are shown in Table 1:

[0085] Table 1 Vehicle parameters

[0086] Vehicle parameters / units value Vehicle mass m / kg 7620 <![CDATA[sprung mass m s / kg]]> 6360 <![CDATA[The moment of inertia I of the sprung mass about the x-axis x / (kg·m -2 )]]> 30782.4 <![CDATA[The moment of inertia I of the sprung mass about the z-axis z / (kg·m -2 )]]> 7695.6 Distance from center of mass to front axle a / m 3.105 Distance from center of mass to rear axle (b / m) 1.385

[0087] The dynamic parameters used in the fixed parameter dynamic model are empirical parameters, specifically the front axle tire cornering stiffness k f 150000N·rad -1 , rear axle tire cornering stiffness k r 460000N·rad -1 , suspension roll stiffness K s 130000 Nm·rad -1 , suspension roll damping C s 30000Nm·rad -1 .

[0088] In this example, the maximum error R between the model output and the TruckSim output is selected. maxAs an indicator for evaluating output results, its calculation formula is:

[0089] R max =|max(y model )-max(y TruckSim )| Formula (17)

[0090] In formula (17), y model Output of fixed parameter dynamic model or adaptive parameter hybrid dynamic model; y TruckSim Output for TruckSim model.

[0091] The speed of the autonomous vehicle is 70 km / h, the steering wheel angle is in the double lane change condition, and the road adhesion coefficient is 0.85. The test results under the above simulation environment are as follows Figure 4 、 Figure 5 and Figure 6 As shown. Figure 4 It can be seen that the center of mass slip angle of the adaptive parameter hybrid dynamics model is closer to the actual vehicle, and the maximum errors of the center of mass slip angle of the fixed parameter dynamics model and the adaptive parameter hybrid dynamics model are 0.0039 rad and 7.84×10 -5 rad; from Figure 5 It can be seen that the yaw rate outputs of the fixed parameter dynamics model and the adaptive parameter hybrid dynamics model are close, and both can accurately reflect the yaw rate changes of the actual vehicle; Figure 6 It can be seen that at the peak of the roll angle change, the fixed parameter dynamics model has a large error with the actual vehicle roll angle change, and cannot accurately reflect the dynamic change of the vehicle roll angle in the nonlinear region; while the output of the adaptive parameter hybrid dynamics model is closer to the actual vehicle roll angle output, and can accurately reflect the actual vehicle roll angle change. The maximum roll angle errors of the two models are 0.0061rad and 0.0016rad respectively. This proves that the GRNN model trained with the SQP algorithm solution can accurately predict the vehicle's roll angle according to the vehicle steering wheel angle δ and longitudinal speed v x The dynamic changes of the vehicle mechanism model are adaptively identified to accurately identify the dynamic parameters, which verifies the effectiveness of the method proposed in this invention.

[0092] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention.

[0093] It should be understood that the present invention is not limited to the above description and that various modifications and changes may be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. An adaptive identification method for dynamic parameters of an autonomous driving vehicle based on SQP and GRNN, characterized by: include: The four parts include the establishment of the autonomous driving vehicle mechanism model, the establishment of the GRNN model, the acquisition of the training set based on the SQP algorithm, and the adaptive identification of parameters based on the trained GRNN model; The establishment of the autonomous driving vehicle mechanism model is specifically as follows: establishing a three-degree-of-freedom model of the autonomous driving vehicle including the vehicle's lateral motion, yaw motion, and roll motion, and its differential equation expression is specifically as follows: In formula (1): v y is the lateral velocity; k f is the front axle tire cornering stiffness; k r is the rear axle tire cornering stiffness; I x is the moment of inertia of the sprung mass around the x-axis; I xs =I x +m s h 2 is the moment of inertia of the suspension sprung mass around the roll axis; m is the vehicle mass; m s is the sprung mass; h is the height from the center of mass to the roll center; δ f is the front wheel angle, which is linearly related to the steering wheel angle δ; v x is the longitudinal velocity; a is the distance from the center of mass to the front axle; b is the distance from the center of mass to the rear axle; ω is the yaw angular velocity; I z is the moment of inertia of the sprung mass around the z-axis; K s is the suspension roll stiffness; C s is the suspension roll damping; φ is the roll angle; g is the acceleration due to gravity; The GRNN model is established by constructing a GRNN model including an input layer, a pattern layer, a summation layer, and an output layer; wherein the input layer parameters include variables that can reflect the driving state of the vehicle, and the output layer parameters include key dynamic parameters in the self-driving car mechanism model; The training set acquisition based on the SQP algorithm is specifically as follows: using the SQP optimization algorithm, the front axle tire cornering stiffness k corresponding to the mechanism model of the autonomous vehicle under typical steering wheel angles and vehicle speeds is obtained. f , rear axle tire cornering stiffness k r and suspension roll stiffness K s , suspension roll damping C s The parameter values ​​of are used as the training data set of the GRNN model; The parameter adaptive identification based on the trained GRNN model is specifically as follows: the operating conditions of the autonomous vehicle and the corresponding SQP solution results are used as the training set of the GRNN model to learn the steering wheel angle δ and longitudinal speed v of the autonomous vehicle. x The nonlinear relationship between the key parameters of the autonomous vehicle mechanism model is that the trained GRNN model can predict the vehicle's steering wheel angle δ and longitudinal speed v x Online identification of the front axle tire cornering stiffness k f , rear axle tire cornering stiffness k r and suspension roll stiffness K s and suspension roll damping C s , to achieve adaptive identification of dynamic parameters of autonomous vehicles.

2. The method for adaptive identification of dynamic parameters of an autonomous driving vehicle based on SQP and GRNN according to claim 1, characterized in that: The solution process of the SQP optimization algorithm is to obtain the key parameter values ​​in the autonomous vehicle mechanism model so that the center of mass slip angle β, yaw rate ω, and roll angle φ output by the autonomous vehicle mechanism model are consistent with the corresponding dynamic indicators of the actual vehicle. The solution formula is: In formula (2), is the i-th center of mass sideslip angle in the time series output by the autonomous vehicle mechanism model; is the sideslip angle of the i-th center of mass in the actual vehicle driving time series; is the i-th yaw rate in the time series output by the autonomous vehicle mechanism model; ω i vehicle is the ith yaw rate in the actual vehicle driving time series; is the i-th roll angle in the time series output by the autonomous vehicle mechanism model; φ i vehicle is the i-th roll angle in the actual vehicle driving time series; α, γ and ξ are weight coefficients respectively.

3. The method for adaptive identification of dynamic parameters of an autonomous driving vehicle based on SQP and GRNN according to claim 2, characterized in that: In order to use the solution of the SQP optimization algorithm as the training set of the GRNN model, the solution process is divided into n time series, and the longitudinal velocity corresponding to the i-th time series under the solution condition is expressed as v xi , the corresponding steering wheel angle is expressed as δ i The corresponding front axle tire cornering stiffness, rear axle tire cornering stiffness, suspension roll stiffness and roll damping are expressed as k fi , k ri , K si , C si ; After solving the problem, we finally get n sets of data sets that can represent the relationship between the steering wheel angle, longitudinal speed and the front axle tire cornering stiffness, rear axle tire cornering stiffness, suspension roll stiffness and roll damping, which are used as the training set of the GRNN model.

4. The method for adaptive identification of dynamic parameters of an autonomous driving vehicle based on SQP and GRNN according to claim 1, characterized in that: The specific process of parameter adaptive identification based on the trained GRNN model is as follows: First, the SQP solution result is used as the training data set of the GRNN model, and the input data is set as the steering wheel angle δ and the longitudinal speed v x , the output data is the front axle tire cornering stiffness k f , rear axle tire cornering stiffness k r and suspension roll stiffness K s and roll damping C s ; Then, the input layer is constructed to determine the dimension of the input data and directly pass the input data to the pattern layer. The pattern layer clusters the input data based on its dimension, uses the Gaussian function to calculate the weight between the input data and each training set sample, multiplies the output data in each training set sample by its corresponding weight, and then the summation layer adds all the results. Finally, the output layer outputs the weighted average of all the results to complete the training of the GRNN model. Finally, the trained GRNN model is trained based on the input data steering wheel angle δ and longitudinal velocity v x , real-time identification and output of the front axle tire cornering stiffness k f , rear axle tire cornering stiffness k r and suspension roll stiffness K s and roll damping C s .

Images