A method and system for dynamic balance adjustment of vehicle chassis attitude
By constructing a recursive state constraint augmented matrix and decomposing a high-order stable subspace, a global robust gain matrix is generated. This solves the problems of large computational load and poor adaptability in the chassis attitude control method of new energy vehicles, achieves effective compensation for multi-source disturbances, and ensures the stability of vehicle attitude.
Patent Information
- Application Number
- CN202611558455.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-09
- Publication Date
- 2026-11-17
AI Technical Summary
Existing chassis attitude control methods suffer from high computational load and high latency when facing multi-source disturbances in new energy vehicles, making it difficult to achieve flexibility and adaptability, and it is also difficult to guarantee control performance under complex working conditions.
By constructing a recursive state constraint augmented matrix, decomposing a high-order stable subspace, performing dimensionality reduction and sparsification, a global robust gain matrix is generated. Combined with Gram matrix eigenvalue decomposition, control commands for the chassis actuator are generated.
It improves the real-time response capability of chassis attitude adjustment, enhances the compensation capability for multi-source disturbances, and ensures the attitude stability and balance of new energy vehicles.
Smart Images

Figure CN123104623A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of adjustment, and in particular relates to a method and system for dynamic balance adjustment of vehicle chassis attitude. Background Technology
[0002] Chassis attitude control systems typically collect vehicle operating state data such as yaw rate, lateral acceleration, and vehicle speed, combine this data with tire mechanical characteristics to construct a model, and use mathematical tools such as control matrix inequalities to solve for the feedback control gain before issuing control commands to the chassis actuators. However, when the vehicle performs emergency obstacle avoidance or faces complex road conditions, the multidimensional state data becomes time-varying, causing the global attitude control matrix inequality to expand in size and exhibit high internal variable correlation. This high-dimensionality and strong correlation characteristic results in a significant computational burden during the online solution process, easily leading to delays in control command calculation.
[0003] In real-world driving environments, new energy vehicles are also affected by multiple external disturbances, including crosswinds, load transfers, and sudden changes in road surface adhesion conditions. These disturbances can cause shifts in the feasible region of tire adhesion. Existing attitude control methods often employ conservative single worst-case assumptions for gain calculation when dealing with such uncertain disturbances. This approach struggles to guarantee the flexibility and adaptability of control performance when faced with complex disturbances. The feasible region boundaries and required compensation gains differ depending on the disturbance state. Relying solely on static conservative models can easily lead to excessive intervention in normal driving conditions or insufficient accuracy in attitude compensation under extreme conditions. Therefore, a novel chassis attitude balance adjustment method is needed that can reasonably reduce and decouple the solution process for high-dimensional control matrix inequalities, while comprehensively evaluating and integrating compensation gains based on multiple disturbance characteristics. Summary of the Invention
[0004] To ensure the stability and balance of new energy vehicles during driving, this invention provides a method and system for dynamic balance adjustment of vehicle chassis posture.
[0005] According to one aspect of the present invention, a method for dynamic balance adjustment of vehicle chassis attitude is provided, the method comprising: The system acquires multi-dimensional state data of vehicle yaw rate, lateral acceleration, vehicle speed, wheel speed and suspension travel. It constructs a time-varying damping operator to transform the lateral acceleration to obtain transient attitude features. It obtains the tire adhesion feasible region based on wheel speed and vehicle speed. It obtains the recursive state constraint augmented matrix based on the transient attitude features and the tire adhesion feasible region. The state sub-blocks of the recursive state constraint augmented matrix are decomposed to obtain a higher-order stable subspace. A decoupling template is obtained based on the higher-order stable subspace. The global attitude control matrix inequality is reduced in dimension and sparsified based on the decoupling template to obtain a structured matrix inequality. Obtain the perturbation vector, and map the perturbation vector to the approximate analytical center point of the perturbed feasible region based on the feasible region boundary mapping function to obtain the candidate attitude compensation gain matrix set corresponding to each perturbation vector; Numerical and vectorization processing is performed on each matrix in the candidate attitude compensation gain matrix set, and the Gram matrix is obtained by inner product. The Gram matrix is decomposed by eigenvalue to obtain the principal eigenvector as the fusion weight vector. The candidate attitude compensation gain matrix set is combined according to the fusion weight vector to obtain the global robust gain matrix. The control command of the chassis actuator is based on the global robust gain matrix, multi-dimensional state data and expected target reference value.
[0006] According to another aspect of the present invention, a vehicle chassis attitude dynamic balance adjustment system is provided, the system comprising the following modules: The construction module is used to obtain multi-dimensional state data of vehicle yaw rate, lateral acceleration, vehicle speed, wheel speed and suspension travel. A time-varying damping operator is constructed to transform the lateral acceleration to obtain transient attitude features. The tire adhesion feasible region is obtained based on wheel speed and vehicle speed. The recursive state constraint augmented matrix is obtained based on the transient attitude features and the tire adhesion feasible region. The decomposition module is used to decompose the state sub-blocks of the recursive state constraint augmented matrix to obtain a higher-order stable subspace. Based on the higher-order stable subspace, a decoupling template is obtained. Based on the decoupling template, the global attitude control matrix inequality is reduced in dimension and sparsified to obtain a structured matrix inequality. The mapping module is used to obtain the perturbation vector. Based on the feasible region boundary mapping function, the perturbation vector is mapped to the analytical center approximation point of the perturbed feasible region to obtain the candidate attitude compensation gain matrix set corresponding to each perturbation vector. The control module is used to perform numerical processing and vectorization on each matrix in the candidate attitude compensation gain matrix set, and obtain the Gram matrix based on the inner product. The Gram matrix is then decomposed into eigenvalues to obtain the principal eigenvectors as fusion weight vectors. The candidate attitude compensation gain matrix set is combined based on the fusion weight vectors to obtain the global robust gain matrix. The control commands for the chassis actuator are based on the global robust gain matrix, multi-dimensional state data, and the desired target reference value.
[0007] This invention integrates vehicle transient attitude features and tire adhesion boundary conditions by constructing a recursive state constraint augmented matrix. It extracts a high-order stable subspace through state sub-block decomposition and constructs a decoupling template, performing dimensionality reduction and sparsification on the global attitude control matrix inequalities, thereby improving solution efficiency and real-time response capability of chassis attitude adjustment. For external uncertainties, it generates a set of candidate attitude compensation gain matrices using feasible region boundary mapping, and extracts fusion weights based on Gram matrices and eigenvalue decomposition to generate a global robust gain matrix. This method achieves compensation for multi-source disturbances, enhances the chassis control's anti-interference capability under disturbed environments, and ensures the attitude stability and balance of new energy vehicles during operation. Attached Figure Description
[0008] Figure 1 A flowchart of a method for dynamic balance adjustment of vehicle chassis posture; Figure 2 A schematic diagram of the feasible region for tire adhesion; Figure 3 This is a schematic diagram of the block structure of the augmented matrix for recursive state constraints. Detailed Implementation
[0009] In this application, the terms "exemplarily" or "for example" are used to indicate that something is being described as an example, illustration, or illustration. Any embodiment or design described as "exemplarily" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplarily" or "for example" is intended to present the relevant concepts in a specific manner.
[0010] Example 1 In Embodiment 1 of the present invention, a method for dynamic balance adjustment of vehicle chassis posture is provided, such as... Figure 1 As shown, it includes: S1. Obtain multi-dimensional state data of vehicle yaw rate, lateral acceleration, vehicle speed, wheel speed and suspension travel. Construct a time-varying damping operator to transform the lateral acceleration to obtain transient attitude features. Obtain the tire adhesion feasible region based on wheel speed and vehicle speed. Obtain the recursive state constraint augmented matrix based on the transient attitude features and the tire adhesion feasible region.
[0011] The vehicle's yaw rate and lateral acceleration are collected in real time by an inertial measurement unit installed at the vehicle's center of gravity. Wheel speed sensors collect the rotational speed of each wheel, retaining the rotational speed as input data for calculating the slip ratio and tire adhesion limit point of each wheel. The arithmetic mean of the rotational speeds of each wheel is calculated as the wheel speed state quantity in the five-dimensional multidimensional state data. The suspension travel of each wheel is collected by a chassis height sensor, retaining the suspension travel of each wheel as input data for estimating the vertical load of each wheel. The arithmetic mean of the suspension travel is calculated as the suspension travel state quantity in the five-dimensional multidimensional state data. The vehicle speed is read. The collected data are synchronized, aligned, and low-pass filtered according to a unified timestamp to form a five-dimensional multidimensional state data column vector containing yaw rate, lateral acceleration, vehicle speed, wheel speed state quantity, and suspension travel state quantity.
[0012] Before constructing the time-varying damping operator, if the current period is the first control cycle, the preset initial gain matrix obtained from offline calibration is read as the gain matrix of the previous control cycle; if the current period is a subsequent control cycle, the global robust gain matrix output by the previous control cycle is used as the gain matrix of the previous control cycle. All singular values of the gain matrix of the previous control cycle are calculated, and the largest singular value is converted into a value greater than 0 and less than 1 through a mapping function, which is then used as the attenuation coefficient of a first-order discrete low-pass filter, thus forming the time-varying damping operator. The lateral acceleration data sequence is input into the time-varying damping operator to obtain the filtered reference lateral acceleration sequence. Then, the input lateral acceleration data sequence is subtracted from the corresponding reference lateral acceleration sequence to obtain the deviation value sequence, which is used as the transient attitude feature.
[0013] For the i-th wheel, the signed real-time wheel slip ratio is calculated by multiplying the wheel speed by the difference between the corresponding wheel rolling radius and the vehicle speed, and then dividing by the maximum value of the vehicle speed and the wheel speed multiplied by the corresponding wheel rolling radius. The suspension travel of the i-th wheel is substituted into a preset load mapping relationship to estimate the vertical load of the i-th wheel. This preset load mapping relationship is determined by the static single-wheel load, suspension stiffness, and calibration correction parameters, and is used to characterize the correspondence between suspension travel and wheel vertical load. Based on the real-time wheel slip ratio of each wheel, and combined with the corresponding wheel vertical load, lateral acceleration, and yaw rate, a preset tire adhesion estimation model is input to estimate the peak adhesion coefficient of the current road surface. The longitudinal tire adhesion limit and lateral tire adhesion limit of each wheel are further calculated. The convex hulls of the longitudinal and lateral adhesion limit points are obtained, and the vertices of the convex hulls are connected in a clockwise order to form a convex polygonal planar region. This convex polygonal planar region is used as the feasible region for tire adhesion, and the discrete boundary coordinate sequence of the feasible region for tire adhesion is extracted.
[0014] The multidimensional state data column vector and the transient attitude feature data vector are arranged together to form the first block matrix of the augmented matrix. The coordinate values of the attached limit points are arranged to form the second block matrix. The historical data of the augmented matrix of the previous control cycle are concatenated column by column at the end of the matrix of the current cycle, and the time window data of the preset length is truncated and retained to generate the recursive state constraint augmented matrix.
[0015] In some implementations, the construction of the time-varying damping operator transforms the lateral acceleration to obtain transient attitude characteristics, including: Calculate all singular values of the gain matrix of the previous control cycle, and convert the largest singular value into a value greater than 0 and less than 1 through a mapping function, and use it as the attenuation coefficient of the first-order discrete low-pass filter to form a time-varying damping operator. The acquired lateral acceleration data sequence is sequentially input into the time-varying damping operator to output the filtered reference lateral acceleration sequence; The deviation value sequence is calculated by subtracting the corresponding reference lateral acceleration sequence from the input lateral acceleration data sequence, and the deviation value sequence is used as the transient attitude feature.
[0016] In the first control cycle, the preset initial gain matrix obtained from offline calibration is read as the gain matrix of the previous control cycle; in subsequent control cycles, the global robust gain matrix output from the previous control cycle is used as the gain matrix of the previous control cycle. Singular values of the gain matrix from the previous control cycle are extracted, and the largest singular value is selected. The value is 10. Input it into the normalized arctangent mapping function. The conversion is performed to obtain 0.875 as the attenuation coefficient. This constitutes the time-varying damping operator of a first-order low-pass filter.
[0017] Lateral acceleration data is acquired with a sampling period of 10ms, and the lateral acceleration value at time k is recorded. Compared with the filter output of the previous cycle Input iterative equation The current filtered reference lateral acceleration value is obtained. .calculate and The difference is 0.175 m / s 2 The deviation value is used as the transient attitude feature of that cycle.
[0018] In some implementations, the process of obtaining the feasible region for tire adhesion based on wheel rotation speed and vehicle speed, and obtaining the recursive state constraint augmented matrix based on transient attitude features and the feasible region for tire adhesion, includes: The effective rolling radius of each wheel is used to convert the corresponding wheel speed into the corresponding wheel linear velocity, and the relative slip ratio data of each wheel linear velocity and vehicle speed is calculated. The vertical load of the corresponding wheel is estimated according to the suspension travel of each wheel and the preset load mapping relationship. The relative slip ratio data, the corresponding wheel vertical load, lateral acceleration and yaw rate are input into the preset tire adhesion estimation model to estimate the peak adhesion coefficient of the current road surface, and the longitudinal and lateral adhesion limit points are mapped in combination with the wheel vertical load. The outer convex hull is obtained for the longitudinal and lateral adhesion limit points, and the vertices of the convex hull are connected in clockwise order. The resulting convex polygonal planar region is defined as the tire adhesion feasible region. The multidimensional state data column vector and the transient attitude feature data vector are arranged together to form the first block matrix of the augmented matrix. The coordinate values of the attached limit points are arranged to form the second block matrix. The historical data of the augmented matrix of the previous control cycle are concatenated column by column at the end of the matrix of the current cycle, and the time window data of the preset length is truncated to form a recursive state constraint augmented matrix.
[0019] Set the effective rolling radius of the wheel to R = 0.32m. Taking the i-th wheel as an example, collect the current wheel speed of that wheel. Given a center-of-gravity longitudinal vehicle speed of 60 rad / s and a center-of-gravity longitudinal vehicle speed of v = 20 m / s, the calculated linear velocity of the wheel is 19.2 m / s. Using the relative slip ratio formula, the current slip ratio of the wheel is -4%. Substituting the wheel suspension travel into the preset load mapping relationship, the vertical load on a single wheel is obtained. =4000N. The current peak adhesion coefficient of the road surface is identified as μ=0.8, based on the single-wheel vertical load. =4000N, which maps the longitudinal adhesion limit points (3200N,0) and (-3200N,0) and the lateral adhesion limit points (0,3200N) and (0,-3200N).
[0020] The rhomboid planar region formed by connecting the four extreme coordinate points clockwise is defined as the feasible region for tire adhesion. The feasible region for tire adhesion is as follows: Figure 2 As shown, the region is a rhomboid convex polygon. The horizontal axis represents the longitudinal force of the tire, and the vertical axis represents the lateral force of the tire. The interior of the region represents the safe and feasible range of tire forces. The five-dimensional multi-dimensional state data column vector and the transient attitude feature vector are arranged together to form the first block matrix of the augmented matrix. The coordinate values of the four adhesion limit points are reshaped into 8-dimensional column vectors and arranged to form the second block matrix. Augmented matrix data are appended column by column to the end of the matrix in the current cycle, and the window data of L=50 steps is truncated to form a recursive state constraint augmented matrix. The recursive state constraint augmented matrix adopts a double-block + sliding time window structure, containing five-dimensional multi-dimensional state data, transient attitude features, and adhesion limit coordinate time sequence information. The matrix structure is as follows. Figure 3As shown, the attachment limit coordinates in the second block matrix are used as boundary constraint parameters in the subsequent construction of the global attitude control matrix inequality.
[0021] S2, decompose the state sub-blocks of the recursive state constraint augmentation matrix to obtain a higher-order stable subspace, obtain a decoupling template based on the higher-order stable subspace, and perform dimensionality reduction and sparsification on the global attitude control matrix inequalities based on the decoupling template to obtain a structured matrix inequality.
[0022] A six-dimensional augmented state time window, composed of five-dimensional multidimensional state data and transient attitude features, is extracted from the first block matrix of the recursive state constraint augmented matrix. A 6×6-dimensional vehicle evolution matrix is identified as a state sub-block based on the six-dimensional augmented state data of adjacent control cycles. This state sub-block matrix is then subjected to Schul decomposition to obtain an orthogonal matrix and an upper triangular matrix. The eigenvalues of the state sub-blocks are determined based on the diagonal elements or diagonal blocks of the upper triangular matrix, and eigenvalues with a modulus less than 1 are identified as stable eigenvalues. Column vectors corresponding to the stable eigenvalues or stable eigenvalue blocks in the orthogonal matrix are extracted to form a basis vector matrix. The transpose of the basis vector matrix, the state sub-blocks, and the basis vector matrix are multiplied sequentially to generate a reduced-dimensional state sub-block matrix within the stable subspace. The basis vector matrix is used to characterize the higher-order stable subspace, and the reduced-dimensional state sub-block matrix is used to characterize the projection result of the state sub-blocks within the higher-order stable subspace.
[0023] Replace the values at the off-diagonal block positions in the dimensionality-reduced state sub-block matrix within the stable subspace that satisfy the preset coupling threshold condition with the value 0 to form a block diagonal structure, and store this block diagonal structure as a decoupling template.
[0024] A global attitude control matrix inequality is established, comprising a state feedback variable matrix and a Lyapunov symmetric positive definite matrix. The matrix terms in this inequality consist of state sub-blocks, control action input coefficients, state deviation weights, control action weights, and boundary constraint parameters formed by the attached limit coordinates in the second block matrix. The global attitude control matrix inequality adopts a discrete-time stability constraint form. Let be a positive definite matrix, where These are the state sub-blocks after dimensionality reduction. Let P be the Lyapunov symmetric positive definite matrix, Y be the state feedback variable matrix, and the corresponding attitude compensation gain matrix be obtained from Y and P. State deviation weights, control action weights, and boundary constraint parameters are used as additional constraint terms in the solution. The control action input coefficients represent the interaction between the control action vector and the vehicle state equation. They are identified from the offline calibration data of the vehicle's two-degree-of-freedom dynamics model, brake actuators, steering actuators, and suspension actuators, and mapped to a higher-order stable subspace according to the reduced-dimensional projection matrix. The state deviation weights and control action weights are... The state feedback variable matrix and the Lyapunov symmetric positive definite matrix are used to limit the attitude deviation and control action amplitude, respectively, as variables to be solved. The pseudo-inverse matrix of the decoupling template is calculated, and a dimension-reduced projection matrix is constructed by combining the basis vector matrix of the higher-order stable subspace with the pseudo-inverse matrix. The dimension-reduced projection matrix and its transpose are multiplied on the left and right by the corresponding matrix of the preset global attitude control matrix inequality, respectively, to generate a dimension-reduced matrix, so that the dimension of the global attitude control matrix inequality is reduced to the dimension of the higher-order stable subspace. Elements in the dimension-reduced matrix whose absolute value is less than the preset sparsification threshold are set to zero to obtain the structured matrix inequality.
[0025] In some implementations, the process of decomposing the state sub-blocks of the recursive state constraint augmentation matrix to obtain a higher-order stable subspace, and obtaining a decoupling template based on the higher-order stable subspace, includes: Extract the five-dimensional multidimensional state data and transient attitude features contained in the first block matrix of the recursive state constraint augmented matrix to form a six-dimensional augmented state time window, and identify the 6×6-dimensional evolution matrix of the corresponding state equation as a state sub-block. The Schul decomposition is completed by obtaining the orthogonal matrix and upper triangular matrix of the state sub-blocks through the Haushold transformation; The eigenvalues of the state sub-blocks are determined based on the diagonal elements or diagonal blocks of the upper triangular matrix, and the eigenvalues with a modulus less than 1 are determined as stable eigenvalues. The column vectors corresponding to the stable eigenvalues or stable eigenvalue blocks in the orthogonal matrix are extracted to form a basis vector matrix. By multiplying the transpose of the basis vector matrix, the state sub-block, and the basis vector matrix in sequence to remove the remaining dimensions, a dimension-reduced state sub-block matrix in the stable subspace is generated. Replace the values of off-diagonal block positions in the dimensionality-reduced state sub-block matrix within the stable subspace whose absolute values are less than the preset coupling threshold with the value 0, and retain the values of off-diagonal block positions whose absolute values are greater than or equal to the preset coupling threshold, generate a block diagonal structure and store it as a decoupling template.
[0026] The data sequences corresponding to multidimensional state data and transient attitude features in the first block matrix of 50 cycles within the time window of the recursive state constraint augmented matrix are extracted, and the 6×6 dimension vehicle evolution matrix A is identified as the state sub-block.
[0027] Applying the Haushold transform to perform the Shure decomposition, the output is an orthogonal matrix Q and an upper triangular matrix T, such that... The condition is met. Extract four stable eigenvalues from T: 0.85, 0.6+0.2i, 0.6-0.2i, and 0.9. Among these, 0.6+0.2i and 0.6-0.2i are the complex conjugate eigenvalues corresponding to the same 2×2 real Shure diagonal block. Extract two columns of real vectors corresponding to this 2×2 diagonal block from Q, and extract the column vectors of real eigenvalues corresponding to 0.85 and 0.9, forming a 6×4 basis vector matrix V. Calculate... Generate a 4×4 dimension high-order stable subspace matrix .
[0028] In the dimensionality-reduced state sub-block matrix within the stable subspace, off-diagonal block elements with absolute values less than a preset coupling threshold are assigned zero. The difference between the matrices before and after zeroing is used as the coupling residual to participate in the additional constraint verification of the structured matrix inequality, generating a block diagonal structure as a decoupling template.
[0029] In some implementations, the dimensionality reduction and sparsification of the global attitude control matrix inequalities based on the decoupling template are performed to obtain structured matrix inequalities, including: Calculate the pseudo-inverse matrix of the decoupling template; Construct a dimension-reduced projection matrix by combining the basis vector matrix and pseudo-inverse matrix of the higher-order stable subspace; The dimension-reduced projection matrix and its transpose are multiplied on the left and right by the preset global attitude control matrix inequality matrix, respectively, to generate a dimension-reduced matrix, thereby reducing the dimension of the global attitude control matrix inequality to the dimension of the higher-order stable subspace. Iterate through each element in the dimension reduction matrix, set the elements whose absolute value is less than the preset sparsity threshold to zero, and construct the structured matrix inequality.
[0030] Obtain a 4×4 dimension decoupling template Find its pseudo-inverse matrix. Multiplying the 6×4 basis vector matrix V by the pseudo-inverse matrix on the right. Calculate and generate a 6×4 dimension-reduced projection matrix G. For the matrix W corresponding to the preset 6×6 dimensional global attitude control matrix inequality, calculate... Generate a 4×4 dimension-reduced matrix The dimensionality reduction matrix is used to solve for control variables in a higher-order stable subspace. During offline calibration, the control variables in the higher-order stable subspace are mapped back to the original state space through the basis vector matrix, while retaining the column positions corresponding to the five-dimensional multi-dimensional state data. First, the 3×6-dimensional attitude compensation gain matrix corresponding to the six-dimensional augmented state is obtained; then, the reduction coefficient is calculated based on the current lateral acceleration, the current reference lateral acceleration, and the current desired lateral acceleration. ,in The current lateral acceleration, As the reference lateral acceleration, Let ε be the desired lateral acceleration, and let ε be a minimal constant to prevent the denominator from being zero. Perform preset upper and lower limit processing; when When the deviation is less than the preset threshold, Set to the preset safety value or the effective value of the previous control cycle; adjust the 3×1 gain column corresponding to the sixth-dimensional transient attitude feature. according to Converted to the lateral acceleration channel, among which This is the 3×1 gain column corresponding to the original lateral acceleration. The synthesized lateral acceleration channel gain column avoids introducing the reference lateral acceleration as an independent control state, resulting in a 3×5 dimensional attitude compensation gain matrix multiplied by the five-dimensional state deviation column vector.
[0031] The 16 elements within the reduced-dimensional matrix are iterated and read, and elements whose absolute values are less than a threshold are selected. Elements such as 0.000085 are assigned the value 0.0, resulting in a structured matrix inequality.
[0032] S3. Obtain the perturbation vector. Based on the feasible region boundary mapping function, map the perturbation vector to the approximate analytical center point of the perturbed feasible region to obtain the candidate attitude compensation gain matrix set corresponding to each perturbation vector.
[0033] By establishing a two-degree-of-freedom vehicle dynamics model to observe the deviation between actual multidimensional state data and theoretical state data, and combining the output data of an external wind resistance observer and road surface roughness data, a multi-source disturbance vector is constructed. Each disturbance component in the multi-source disturbance vector is normalized according to its corresponding preset reference scale to obtain a disturbance column vector; the subsequent feasible region boundary mapping function is calculated using this disturbance column vector as the independent variable.
[0034] Multiple analytical center sample points corresponding to disturbance samples are pre-calculated based on feasible region boundary constraints. A third-order Taylor expansion multivariate polynomial is fitted based on the analytical center sample points, and this multivariate polynomial is used as the feasible region boundary mapping function. During online calculation, the disturbance column vector is substituted as the independent variable into the feasible region boundary mapping function to obtain the analytical center approximation point of the disturbed feasible region. The analytical center approximation point is then verified based on the boundary constraints of the disturbed feasible region. If the analytical center approximation point exceeds the disturbed feasible region, it is projected to the nearest feasible point inside the boundary of the disturbed feasible region. Multiple disturbance vectors are composed of the disturbance column vector of the current control cycle and the historical disturbance column vector of a preset length.
[0035] The coordinates of the approximate analytical center point are used as boundary compensation parameters, and the coordinate components of the approximate analytical center point and their second-order and third-order combinations are arranged in a fixed order as a coordinate combination column vector. The coordinate combination column vector consists of the coordinates of the approximate analytical center point and their product terms. The pre-stored constant parameter block corresponding to the structured matrix inequality is called to perform matrix multiplication and weighted summation with the coordinate combination column vector to calculate the candidate attitude compensation gain matrix.
[0036] In the offline calibration process, the structured matrix inequalities are first solved jointly for multiple analytical center sample points to determine a Lyapunov symmetric positive definite matrix P that is applicable to the multiple analytical center sample points. Then, under the condition of fixing the Lyapunov symmetric positive definite matrix P, the state feedback variable matrix corresponding to each analytical center sample point is solved separately. , and by The corresponding sample attitude compensation gain matrix is obtained, and then numerical fitting is performed using the coordinate components of the analytical center sample point as input and the matrix elements of the sample attitude compensation gain matrix as output to obtain the pre-stored constant parameter block. During offline calibration, the matrix norm error between the numerical fitting output matrix and the corresponding sample attitude compensation gain matrix is calculated, and the matrix norm error is limited to a preset error upper limit. After calculating the candidate attitude compensation gain matrix, it is re-substituted into the structured matrix inequality for verification. After the verification is passed, it is included in the candidate attitude compensation gain matrix set. Each sample attitude compensation gain matrix satisfies the same structured matrix inequality and the same variable constraints during offline calibration. Therefore, the candidate attitude compensation gain matrix calculated by the pre-stored constant parameter block is used to maintain the stable constraint relationship within the corresponding disturbance sample range. The pre-stored constant parameter block is used to characterize the correspondence between the analytical center coordinates and the candidate attitude compensation gain matrix. The analytical center approximation points corresponding to all disturbance column vectors are traversed and the above calculation process is repeated to collect all the calculated candidate attitude compensation gain matrices to obtain the candidate attitude compensation gain matrix set.
[0037] In some implementations, the perturbation vector is mapped to an approximate analytical center point of the perturbed feasible region based on the feasible region boundary mapping function, resulting in a set of candidate attitude compensation gain matrices for each perturbation vector, including: Each perturbation vector is normalized according to the corresponding preset reference scale to obtain each perturbation column vector. Each processed perturbation column vector is then substituted as an independent variable into the offline stored third-order Taylor expansion multivariate polynomial for numerical calculation to generate a set of coordinate vectors. The multivariate polynomial is used as the feasible region boundary mapping function to realize the projection operation. The output set of coordinate vectors is defined as the analytical center approximation point of the perturbed feasible region; The coordinate values of each analytical center approximation point and their second-order and third-order combinations are arranged into a coordinate combination column vector. The coordinate combination column vector is then multiplied and weighted by matrix multiplication with the pre-stored constant parameter block of the structured matrix inequality to calculate the determined candidate attitude compensation gain matrix. All generated candidate attitude compensation gain matrices are aggregated into a candidate attitude compensation gain matrix set.
[0038] Obtain the normalized perturbation column vector Substituting it into a third-order Taylor series expansion of a multivariate polynomial It is calculated as a feasible region boundary mapping function, where l represents the output coordinate component index, and p, q, and r represent the component indices in the perturbation column vector d. and All coefficients are obtained through offline fitting, and the output is a two-dimensional coordinate vector. It is defined as the approximate analytical center point of the perturbed feasible region.
[0039] Based on the coordinate components of the approximate center point and their second- and third-order combinations, the corresponding constant parameter matrix blocks obtained through offline calibration are extracted. These blocks are then weighted and summed according to the element correspondence obtained from offline fitting to obtain a determined 3×5-dimensional candidate attitude compensation gain matrix. N=10 disturbance samples are selected from the disturbance column vectors corresponding to the current control cycle and historical cycles. This calculation is then performed iteratively on each sample, and the resulting set of candidate attitude compensation gain matrices containing 10 elements is output.
[0040] S4. Numerical and vectorization processing is performed on each matrix in the candidate attitude compensation gain matrix set, and the Gram matrix is obtained by inner product. The Gram matrix is decomposed into eigenvalues to obtain the principal eigenvectors as fusion weight vectors. The candidate attitude compensation gain matrix set is combined according to the fusion weight vectors to obtain the global robust gain matrix. The control commands of the chassis actuator are based on the global robust gain matrix, multi-dimensional state data and expected target reference values.
[0041] Each state feedback control gain matrix in the candidate attitude compensation gain matrix set is first numerically processed according to the limiting value of each control output component and the reference scale of each state input component, so that different control channels and state channels are at comparable numerical scales. Then, all rows are expanded into one-dimensional column vectors in order from top to bottom and from left to right. The standard inner product between each pair of one-dimensional column vectors is calculated. The numerical results of each standard inner product are arranged according to the corresponding matrix number as row and column numbers to generate a symmetric and positive semi-definite Gram matrix.
[0042] The characteristic equation of the Gram matrix is calculated, and all eigenvalues are obtained. The eigenvalue with the largest value is extracted. When the largest eigenvalue corresponds to a single eigenvector, the normalized column vector of the eigenvector is calculated. The absolute value of each element is taken and divided by the sum of the absolute values of all elements to obtain a non-negative fusion weight vector with a sum of 1. When the largest eigenvalue corresponds to multiple eigenvectors, the multiple eigenvectors are normalized and averaged separately. The absolute value of each element in the average result is taken and divided by the sum of the absolute values of all elements to obtain a non-negative fusion weight vector with a sum of 1. Taking the absolute value of the eigenvector is only used to obtain the non-negative fusion ratio of each candidate attitude compensation gain matrix. The positive and negative directions of each control channel in the candidate attitude compensation gain matrix remain unchanged. When the signs of corresponding elements in the candidate attitude compensation gain matrices are opposite, the elements of the original matrix participate in the weighted summation, and the control direction is reflected by the sign of the matrix element itself.
[0043] Each scalar element in the fusion weight vector is extracted as a weighting coefficient for the corresponding candidate attitude compensation gain matrix. Each matrix in the candidate attitude compensation gain matrix set is multiplied by its corresponding weighting coefficient, and then summed using matrix addition. The resulting sum is the global robust gain matrix. Since each candidate attitude compensation gain matrix is generated by the same structured matrix inequality and the same variable constraints, and since each candidate attitude compensation gain matrix is calibrated offline under the condition of fixing the same Lyapunov symmetric positive definite matrix P, the corresponding state feedback variable matrix is obtained by solving. , and can be represented as The weighted combination corresponding to the non-negative fusion weight vector with an element sum of 1 satisfies Therefore, the weighted combination is re-substituted into the structured matrix inequality under the same Lyapunov symmetric positive definite matrix P for verification; after the stability constraint is satisfied, it is used to output control commands; if the verification is not satisfied, the candidate attitude compensation gain matrix that has passed the verification is selected as the gain matrix of the current control cycle.
[0044] Based on the steering angle collected by the vehicle steering wheel angle sensor and the pedal opening collected by the pedal displacement sensor, the pedal opening is first input into a preset longitudinal target vehicle speed mapping relationship to obtain the desired vehicle speed. The preset longitudinal target vehicle speed mapping relationship is obtained by offline calibration of the pedal opening, current vehicle speed, and vehicle longitudinal acceleration limit, and is used to characterize the driver's longitudinal speed intention. Then, the Ackerman steering geometry model is used to calculate the desired yaw rate based on the steering wheel angle and the desired vehicle speed, calculate the desired wheel speed based on the desired vehicle speed and the effective rolling radius of the wheel, calculate the desired lateral acceleration based on the desired yaw rate and the desired vehicle speed, and use the preset suspension center travel as the desired suspension travel. This forms the desired target reference value vector corresponding to the multi-dimensional state data. The Ackerman steering geometry model determines the equivalent steering angle of the front wheels based on the vehicle wheelbase, steering wheel angle, and steering transmission ratio, and calculates the desired yaw rate based on the equivalent steering angle of the front wheels and the desired vehicle speed.
[0045] The state deviation vector is obtained by subtracting the column vector of the actual collected multidimensional state data from the vector of the desired target reference value. The global robust gain matrix is then multiplied by the state deviation vector to calculate the control action vector, which includes the desired brake hydraulic pressure, the desired steering angle, and the desired suspension adjustment force.
[0046] The desired brake hydraulic pressure, desired steering angle, and desired suspension adjustment force are each limited and encapsulated into control commands according to the communication protocol of the controller local area network. These commands are then sent to the brake actuator, steering actuator, and suspension actuator, respectively, as control commands for the chassis actuators to maintain attitude balance.
[0047] In some implementations, the step of performing numerical processing and vectorization on each matrix in the candidate attitude compensation gain matrix set and obtaining the Gram matrix based on the inner product includes: Numerical processing is performed on each candidate attitude compensation gain matrix in the candidate attitude compensation gain matrix set according to the limiting value of each control output component and the reference scale of each state input component. Each candidate attitude compensation gain matrix after numerical processing is expanded into a one-dimensional column vector in order from top to bottom and from left to right, thus completing the vectorization process. All one-dimensional column vectors generated by the process are paired and extracted in a loop. The first column vector is transposed and multiplied with the second column vector to obtain the scalar inner product value. Based on the row and column positions of the two-dimensional array corresponding to the order of the extracted vectors, all the calculated inner product values are written into a new symmetric real square matrix as the Gram matrix.
[0048] The candidate attitude compensation gain matrix of each 3×5 dimension in the set is extracted sequentially. After numerical processing according to the limiting value of each control output component and the reference scale of each state input component, it is expanded into a 15×1 dimension one-dimensional column vector in the order from top to bottom and from left to right.
[0049] After the loop completes, the included... to Ten one-dimensional column vectors are vectorized. The inner product of the i-th and j-th column vectors is calculated to obtain a scalar value, such as 15.6. According to the corresponding row and column positions, all the calculated inner product values are written into the corresponding positions to form a 10×10 symmetric real matrix as the Gram matrix.
[0050] In some implementations, the eigenvalue decomposition of the Gram matrix to obtain the principal eigenvector is used as the fusion weight vector. The global robust gain matrix is then obtained by combining the candidate pose compensation gain matrix set based on the fusion weight vector. This includes: Perform Jacobi iteration on the Gram matrix to obtain all eigenvalues and corresponding eigenvectors of the matrix; Compare all calculated eigenvalues to find the positive real eigenvalue with the largest value. When the maximum positive real eigenvalue corresponds to a single eigenvector, calculate the normalized column vector of the eigenvector, take the absolute value of each element and divide it by the sum of the absolute values of all elements to obtain the fusion weight vector. When the maximum positive real eigenvalue corresponds to multiple eigenvectors, normalize and average the multiple eigenvectors respectively, take the absolute value of each element in the average result and divide it by the sum of the absolute values of all elements to obtain the fusion weight vector. Each scalar element in the fusion weight vector is multiplied by the original candidate attitude compensation gain matrix at the corresponding position in the candidate attitude compensation gain matrix set. All the multiplication results are summed element by element to output the global robust gain matrix.
[0051] For a 10×10 dimensional Gram matrix, Jacobi iteration is used to obtain 10 real eigenvalues and their corresponding eigenvectors. The largest positive real eigenvalue, 142.5, is found, and the normalized unit eigenvector corresponding to this eigenvalue is calculated. The absolute value is then divided by the sum of the absolute values of all elements to generate a 10-dimensional non-negative fusion weight vector w=[0.12,0.23,0.05,...]. The above absolute value and normalization process is used to obtain the non-negative fusion ratio of the candidate attitude compensation gain matrix. The positive and negative elements in each candidate attitude compensation gain matrix remain unchanged, and the control direction is still determined by the candidate attitude compensation gain matrix itself participating in the weighted summation.
[0052] Extract the weight elements from the vector Multiply by the 3×5 dimensional candidate pose compensation gain matrix at the corresponding position. The elements of all product result matrices are summed and accumulated to output a 3×5 dimension global robust gain matrix.
[0053] In some implementations, the control commands for the chassis actuator based on the global robust gain matrix, multidimensional state data, and desired target reference values include: The state deviation is obtained by subtracting the corresponding expected target reference value from the real-time acquired multi-dimensional state data of vehicle yaw rate, lateral acceleration, vehicle speed, wheel speed and suspension travel. The state deviations are then vertically concatenated into the current state deviation column vector. The control action vector is obtained by calculating the matrix product between the global robustness gain matrix and the state deviation column vector. The first parameter component in the control motion vector is input into the amplitude limiting calculation module, which outputs the desired braking hydraulic pressure and encapsulates it into a control command to be sent to the brake actuator. The second parameter component of the control action vector is limited to generate the desired steering angle, which is then encapsulated into a control command and sent to the steering actuator. The third parameter component of the control action vector is limited to generate the desired suspension adjustment force, which is then encapsulated into a control command and sent to the suspension actuator.
[0054] Obtain the vehicle's yaw rate (0.5 rad / s) and lateral acceleration. The vehicle speed is 20 m / s, the arithmetic mean of the rotational speeds of all wheels is 62 rad / s, and the arithmetic mean of the suspension travel is 0.015 m. A target reference value, such as a yaw rate of 0.45 rad / s, is obtained. The state deviations, such as a yaw rate deviation of 0.05 rad / s, are calculated and subtracted. These five scalar state deviations are then longitudinally concatenated to form a 5×1 dimension state deviation column vector e.
[0055] Using a 3×5 global robust gain matrix Multiplying the state deviation column vector e by the left yields a 3×1 dimension control action vector. . Componentize the first parameter For example, the 3.5 input limiting calculation module outputs the desired braking hydraulic pressure of 3.5MPa and encapsulates it into a command to be sent to the brake actuator; the second parameter component The desired steering angle is limited and encapsulated into a command, which is then sent to the steering actuator; the third parameter component is... The desired suspension adjustment force is limited and encapsulated into a command, which is then sent to the suspension actuator.
[0056] Example 2 Embodiment 2 of the present invention proposes a vehicle chassis attitude dynamic balance adjustment system, comprising the following modules: The construction module is used to obtain multi-dimensional state data of vehicle yaw rate, lateral acceleration, vehicle speed, wheel speed and suspension travel. A time-varying damping operator is constructed to transform the lateral acceleration to obtain transient attitude features. The tire adhesion feasible region is obtained based on wheel speed and vehicle speed. The recursive state constraint augmented matrix is obtained based on the transient attitude features and the tire adhesion feasible region. The decomposition module is used to decompose the state sub-blocks of the recursive state constraint augmented matrix to obtain a higher-order stable subspace. Based on the higher-order stable subspace, a decoupling template is obtained. Based on the decoupling template, the global attitude control matrix inequality is reduced in dimension and sparsified to obtain a structured matrix inequality. The mapping module is used to obtain the perturbation vector. Based on the feasible region boundary mapping function, the perturbation vector is mapped to the analytical center approximation point of the perturbed feasible region to obtain the candidate attitude compensation gain matrix set corresponding to each perturbation vector. The control module is used to perform numerical processing and vectorization on each matrix in the candidate attitude compensation gain matrix set, and obtain the Gram matrix based on the inner product. The Gram matrix is then decomposed into eigenvalues to obtain the principal eigenvectors as fusion weight vectors. The candidate attitude compensation gain matrix set is combined based on the fusion weight vectors to obtain the global robust gain matrix. The control commands for the chassis actuator are based on the global robust gain matrix, multi-dimensional state data, and the desired target reference value.
[0057] In some implementations, the construction of the time-varying damping operator transforms the lateral acceleration to obtain transient attitude characteristics, including: Calculate all singular values of the gain matrix of the previous control cycle, and convert the largest singular value into a value greater than 0 and less than 1 through a mapping function, and use it as the attenuation coefficient of the first-order discrete low-pass filter to form a time-varying damping operator. The acquired lateral acceleration data sequence is sequentially input into the time-varying damping operator to output the filtered reference lateral acceleration sequence; The deviation value sequence is calculated by subtracting the corresponding reference lateral acceleration sequence from the input lateral acceleration data sequence, and the deviation value sequence is used as the transient attitude feature.
[0058] In some implementations, the process of obtaining the feasible region for tire adhesion based on wheel rotation speed and vehicle speed, and obtaining the recursive state constraint augmented matrix based on transient attitude features and the feasible region for tire adhesion, includes: The effective rolling radius of each wheel is used to convert the corresponding wheel speed into the corresponding wheel linear velocity, and the relative slip ratio data of each wheel linear velocity and vehicle speed is calculated. The vertical load of the corresponding wheel is estimated according to the suspension travel of each wheel and the preset load mapping relationship. The relative slip ratio data, the corresponding wheel vertical load, lateral acceleration and yaw rate are input into the preset tire adhesion estimation model to estimate the peak adhesion coefficient of the current road surface, and the longitudinal and lateral adhesion limit points are mapped in combination with the wheel vertical load. The outer convex hull is obtained for the longitudinal and lateral adhesion limit points, and the vertices of the convex hull are connected in clockwise order. The resulting convex polygonal planar region is defined as the tire adhesion feasible region. The multidimensional state data column vector and the transient attitude feature data vector are arranged together to form the first block matrix of the augmented matrix. The coordinate values of the attached limit points are arranged to form the second block matrix. The historical data of the augmented matrix of the previous control cycle are concatenated column by column at the end of the matrix of the current cycle, and the time window data of the preset length is truncated to form a recursive state constraint augmented matrix.
[0059] In some implementations, the process of decomposing the state sub-blocks of the recursive state constraint augmentation matrix to obtain a higher-order stable subspace, and obtaining a decoupling template based on the higher-order stable subspace, includes: Extract the five-dimensional multidimensional state data and transient attitude features contained in the first block matrix of the recursive state constraint augmented matrix to form a six-dimensional augmented state time window, and identify the 6×6-dimensional evolution matrix of the corresponding state equation as a state sub-block. The Schul decomposition is completed by obtaining the orthogonal matrix and upper triangular matrix of the state sub-blocks through the Haushold transformation; The eigenvalues of the state sub-blocks are determined based on the diagonal elements or diagonal blocks of the upper triangular matrix, and the eigenvalues with a modulus less than 1 are determined as stable eigenvalues. The column vectors corresponding to the stable eigenvalues or stable eigenvalue blocks in the orthogonal matrix are extracted to form a basis vector matrix. By multiplying the transpose of the basis vector matrix, the state sub-block, and the basis vector matrix in sequence to remove the remaining dimensions, a dimension-reduced state sub-block matrix in the stable subspace is generated. Replace the values of off-diagonal block positions in the dimensionality-reduced state sub-block matrix within the stable subspace whose absolute values are less than the preset coupling threshold with the value 0, and retain the values of off-diagonal block positions whose absolute values are greater than or equal to the preset coupling threshold, generate a block diagonal structure and store it as a decoupling template.
[0060] In some implementations, the dimensionality reduction and sparsification of the global attitude control matrix inequalities based on the decoupling template are performed to obtain structured matrix inequalities, including: Calculate the pseudo-inverse matrix of the decoupling template; Construct a dimension-reduced projection matrix by combining the basis vector matrix and pseudo-inverse matrix of the higher-order stable subspace; The dimension-reduced projection matrix and its transpose are multiplied on the left and right by the preset global attitude control matrix inequality matrix, respectively, to generate a dimension-reduced matrix, thereby reducing the dimension of the global attitude control matrix inequality to the dimension of the higher-order stable subspace. Iterate through each element in the dimension reduction matrix, set the elements whose absolute value is less than the preset sparsity threshold to zero, and construct the structured matrix inequality.
[0061] In some implementations, the perturbation vector is mapped to an approximate analytical center point of the perturbed feasible region based on the feasible region boundary mapping function, resulting in a set of candidate attitude compensation gain matrices for each perturbation vector, including: Each perturbation vector is normalized according to the corresponding preset reference scale to obtain each perturbation column vector. Each processed perturbation column vector is then substituted as an independent variable into the offline stored third-order Taylor expansion multivariate polynomial for numerical calculation to generate a set of coordinate vectors. The multivariate polynomial is used as the feasible region boundary mapping function to realize the projection operation. The output set of coordinate vectors is defined as the analytical center approximation point of the perturbed feasible region; The coordinate values of each analytical center approximation point and their second-order and third-order combinations are arranged into a coordinate combination column vector. The coordinate combination column vector is then multiplied and weighted by matrix multiplication with the pre-stored constant parameter block of the structured matrix inequality to calculate the determined candidate attitude compensation gain matrix. All generated candidate attitude compensation gain matrices are aggregated into a candidate attitude compensation gain matrix set.
[0062] In some implementations, the step of performing numerical processing and vectorization on each matrix in the candidate attitude compensation gain matrix set and obtaining the Gram matrix based on the inner product includes: Numerical processing is performed on each candidate attitude compensation gain matrix in the candidate attitude compensation gain matrix set according to the limiting value of each control output component and the reference scale of each state input component. Each candidate attitude compensation gain matrix after numerical processing is expanded into a one-dimensional column vector in order from top to bottom and from left to right, thus completing the vectorization process. All one-dimensional column vectors generated by the process are paired and extracted in a loop. The first column vector is transposed and multiplied with the second column vector to obtain the scalar inner product value. Based on the row and column positions of the two-dimensional array corresponding to the order of the extracted vectors, all the calculated inner product values are written into a new symmetric real square matrix as the Gram matrix.
[0063] In some implementations, the eigenvalue decomposition of the Gram matrix to obtain the principal eigenvector is used as the fusion weight vector. The global robust gain matrix is then obtained by combining the candidate pose compensation gain matrix set based on the fusion weight vector. This includes: Perform Jacobi iteration on the Gram matrix to obtain all eigenvalues and corresponding eigenvectors of the matrix; Compare all calculated eigenvalues to find the positive real eigenvalue with the largest value. When the maximum positive real eigenvalue corresponds to a single eigenvector, calculate the normalized column vector of the eigenvector, take the absolute value of each element and divide it by the sum of the absolute values of all elements to obtain the fusion weight vector. When the maximum positive real eigenvalue corresponds to multiple eigenvectors, normalize and average the multiple eigenvectors respectively, take the absolute value of each element in the average result and divide it by the sum of the absolute values of all elements to obtain the fusion weight vector. Each scalar element in the fusion weight vector is multiplied by the original candidate attitude compensation gain matrix at the corresponding position in the candidate attitude compensation gain matrix set. All the multiplication results are summed element by element to output the global robust gain matrix.
[0064] In some implementations, the control commands for the chassis actuator based on the global robust gain matrix, multidimensional state data, and desired target reference values include: The state deviation is obtained by subtracting the corresponding expected target reference value from the real-time acquired multi-dimensional state data of vehicle yaw rate, lateral acceleration, vehicle speed, wheel speed and suspension travel. The state deviations are then vertically concatenated into the current state deviation column vector. The control action vector is obtained by calculating the matrix product between the global robustness gain matrix and the state deviation column vector. The first parameter component in the control motion vector is input into the amplitude limiting calculation module, which outputs the desired braking hydraulic pressure and encapsulates it into a control command to be sent to the brake actuator. The second parameter component of the control action vector is limited to generate the desired steering angle, which is then encapsulated into a control command and sent to the steering actuator. The third parameter component of the control action vector is limited to generate the desired suspension adjustment force, which is then encapsulated into a control command and sent to the suspension actuator.
[0065] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for dynamic balance adjustment of vehicle chassis attitude, characterized in that, include: The system acquires multi-dimensional state data of vehicle yaw rate, lateral acceleration, vehicle speed, wheel speed and suspension travel. It constructs a time-varying damping operator to transform the lateral acceleration to obtain transient attitude features. It obtains the tire adhesion feasible region based on wheel speed and vehicle speed. It obtains the recursive state constraint augmented matrix based on the transient attitude features and the tire adhesion feasible region. The state sub-blocks of the recursive state constraint augmented matrix are decomposed to obtain a higher-order stable subspace. A decoupling template is obtained based on the higher-order stable subspace. The global attitude control matrix inequality is reduced in dimension and sparsified based on the decoupling template to obtain a structured matrix inequality. Obtain the perturbation vector, and map the perturbation vector to the approximate analytical center point of the perturbed feasible region based on the feasible region boundary mapping function to obtain the candidate attitude compensation gain matrix set corresponding to each perturbation vector; Numerical and vectorization processing is performed on each matrix in the candidate attitude compensation gain matrix set, and the Gram matrix is obtained by inner product. The Gram matrix is decomposed by eigenvalue to obtain the principal eigenvector as the fusion weight vector. The candidate attitude compensation gain matrix set is combined according to the fusion weight vector to obtain the global robust gain matrix. The control command of the chassis actuator is based on the global robust gain matrix, multi-dimensional state data and expected target reference value.
2. The method according to claim 1, characterized in that, The constructed time-varying damping operator transforms the lateral acceleration to obtain transient attitude characteristics, including: Calculate all singular values of the gain matrix of the previous control cycle, and convert the largest singular value into a value greater than 0 and less than 1 through a mapping function, and use it as the attenuation coefficient of the first-order discrete low-pass filter to form a time-varying damping operator. The acquired lateral acceleration data sequence is sequentially input into the time-varying damping operator to output the filtered reference lateral acceleration sequence; The deviation value sequence is calculated by subtracting the corresponding reference lateral acceleration sequence from the input lateral acceleration data sequence, and the deviation value sequence is used as the transient attitude feature.
3. The method according to claim 1, characterized in that, The feasible region for tire adhesion is obtained based on wheel rotation speed and vehicle speed. A recursive state constraint augmented matrix is then derived based on transient attitude characteristics and the feasible region for tire adhesion, including: The effective rolling radius of each wheel is used to convert the corresponding wheel speed into the corresponding wheel linear velocity, and the relative slip ratio data of each wheel linear velocity and vehicle speed is calculated. The vertical load of the corresponding wheel is estimated according to the suspension travel of each wheel and the preset load mapping relationship. The relative slip ratio data, the corresponding wheel vertical load, lateral acceleration and yaw rate are input into the preset tire adhesion estimation model to estimate the peak adhesion coefficient of the current road surface, and the longitudinal and lateral adhesion limit points are mapped in combination with the wheel vertical load. The outer convex hull is obtained for the longitudinal and lateral adhesion limit points, and the vertices of the convex hull are connected in clockwise order. The resulting convex polygonal planar region is defined as the feasible region for tire adhesion. The multidimensional state data column vector and the transient attitude feature data vector are arranged together to form the first block matrix of the augmented matrix. The coordinate values of the attached limit points are arranged to form the second block matrix. The historical data of the augmented matrix of the previous control cycle are concatenated column by column at the end of the matrix of the current cycle, and the time window data of the preset length is truncated to form a recursive state constraint augmented matrix.
4. The method according to claim 1, characterized in that, The process of decomposing the state sub-blocks of the recursive state constraint augmentation matrix to obtain a higher-order stable subspace, and obtaining a decoupling template based on the higher-order stable subspace, includes: Extract the five-dimensional multidimensional state data and transient attitude features contained in the first block matrix of the recursive state constraint augmented matrix to form a six-dimensional augmented state time window, and identify the 6×6-dimensional evolution matrix of the corresponding state equation as a state sub-block. The Schul decomposition is completed by obtaining the orthogonal matrix and upper triangular matrix of the state sub-blocks through the Haushold transformation; The eigenvalues of the state sub-blocks are determined based on the diagonal elements or diagonal blocks of the upper triangular matrix, and the eigenvalues with a modulus less than 1 are determined as stable eigenvalues. The column vectors corresponding to the stable eigenvalues or stable eigenvalue blocks in the orthogonal matrix are extracted to form a basis vector matrix. By multiplying the transpose of the basis vector matrix, the state sub-block, and the basis vector matrix in sequence to remove the remaining dimensions, a dimension-reduced state sub-block matrix in the stable subspace is generated. Replace the values of off-diagonal block positions in the dimensionality-reduced state sub-block matrix within the stable subspace whose absolute values are less than the preset coupling threshold with the value 0, and retain the values of off-diagonal block positions whose absolute values are greater than or equal to the preset coupling threshold, generate a block diagonal structure and store it as a decoupling template.
5. The method according to claim 2, characterized in that, The method of reducing and sparsifying the global attitude control matrix inequalities based on decoupling templates to obtain structured matrix inequalities includes: Calculate the pseudo-inverse matrix of the decoupling template; Construct a dimension-reduced projection matrix by combining the basis vector matrix and pseudo-inverse matrix of the higher-order stable subspace; The dimension-reduced projection matrix and its transpose are multiplied on the left and right by the preset global attitude control matrix inequality matrix, respectively, to generate a dimension-reduced matrix, thereby reducing the dimension of the global attitude control matrix inequality to the dimension of the higher-order stable subspace. Iterate through each element in the dimension reduction matrix, set the elements whose absolute value is less than the preset sparsity threshold to zero, and construct the structured matrix inequality.
6. The method according to claim 1, characterized in that, The method of mapping the perturbation vector to the approximate analytical center point of the perturbation feasible region based on the feasible region boundary mapping function yields a set of candidate attitude compensation gain matrices for each perturbation vector, including: Each perturbation vector is normalized according to the corresponding preset reference scale to obtain each perturbation column vector. Each processed perturbation column vector is then substituted as an independent variable into the offline stored third-order Taylor expansion multivariate polynomial for numerical calculation to generate a set of coordinate vectors. The multivariate polynomial is used as the feasible region boundary mapping function to realize the projection operation. The output set of coordinate vectors is defined as the analytical center approximation point of the perturbed feasible region; The coordinate values of each analytical center approximation point and their second-order and third-order combinations are arranged into a coordinate combination column vector. The coordinate combination column vector is then multiplied and weighted by matrix multiplication with the pre-stored constant parameter block of the structured matrix inequality to calculate the determined candidate attitude compensation gain matrix. All generated candidate attitude compensation gain matrices are aggregated into a candidate attitude compensation gain matrix set.
7. The method according to claim 1, characterized in that, The step of numerically processing and vectorizing each matrix in the candidate attitude compensation gain matrix set, and obtaining the Gram matrix based on the inner product, includes: Numerical processing is performed on each candidate attitude compensation gain matrix in the candidate attitude compensation gain matrix set according to the limiting value of each control output component and the reference scale of each state input component. Each candidate attitude compensation gain matrix after numerical processing is expanded into a one-dimensional column vector in order from top to bottom and from left to right, thus completing the vectorization process. All one-dimensional column vectors generated by the process are paired and extracted in a loop. The first column vector is transposed and multiplied with the second column vector to obtain the scalar inner product value. Based on the row and column positions of the two-dimensional array corresponding to the order of the extracted vectors, all the calculated inner product values are written into a new symmetric real square matrix as the Gram matrix.
8. The method according to claim 1, characterized in that, The eigenvalue decomposition of the Gram matrix yields the principal eigenvectors, which are then used as fusion weight vectors. These fusion weight vectors are then used to combine the candidate pose compensation gain matrix set to obtain the global robust gain matrix, including: Perform Jacobi iteration on the Gram matrix to obtain all eigenvalues and corresponding eigenvectors of the matrix; Compare all calculated eigenvalues to find the positive real eigenvalue with the largest value. When the maximum positive real eigenvalue corresponds to a single eigenvector, calculate the normalized column vector of the eigenvector, take the absolute value of each element and divide it by the sum of the absolute values of all elements to obtain the fusion weight vector. When the maximum positive real eigenvalue corresponds to multiple eigenvectors, normalize and average the multiple eigenvectors respectively, take the absolute value of each element in the average result and divide it by the sum of the absolute values of all elements to obtain the fusion weight vector. Each scalar element in the fusion weight vector is multiplied by the original candidate attitude compensation gain matrix at the corresponding position in the candidate attitude compensation gain matrix set. All the multiplication results are summed element by element to output the global robust gain matrix.
9. The method according to claim 1, characterized in that, The control commands for the chassis actuator based on the global robust gain matrix, multi-dimensional state data, and desired target reference value include: The state deviation is obtained by subtracting the corresponding expected target reference value from the real-time acquired multi-dimensional state data of vehicle yaw rate, lateral acceleration, vehicle speed, wheel speed and suspension travel. The state deviations are then vertically concatenated into the current state deviation column vector. The control action vector is obtained by calculating the matrix product between the global robustness gain matrix and the state deviation column vector. The first parameter component in the control motion vector is input into the amplitude limiting calculation module, which outputs the desired braking hydraulic pressure and encapsulates it into a control command to be sent to the brake actuator. The second parameter component of the control action vector is limited to generate the desired steering angle, which is then encapsulated into a control command and sent to the steering actuator. The third parameter component of the control action vector is limited to generate the desired suspension adjustment force, which is then encapsulated into a control command and sent to the suspension actuator.
10. A vehicle chassis attitude dynamic balance adjustment system, characterized in that, include: The construction module is used to obtain multi-dimensional state data of vehicle yaw rate, lateral acceleration, vehicle speed, wheel speed and suspension travel. A time-varying damping operator is constructed to transform the lateral acceleration to obtain transient attitude features. The tire adhesion feasible region is obtained based on wheel speed and vehicle speed. The recursive state constraint augmented matrix is obtained based on the transient attitude features and the tire adhesion feasible region. The decomposition module is used to decompose the state sub-blocks of the recursive state constraint augmented matrix to obtain a higher-order stable subspace. Based on the higher-order stable subspace, a decoupling template is obtained. Based on the decoupling template, the global attitude control matrix inequality is reduced in dimension and sparsified to obtain a structured matrix inequality. The mapping module is used to obtain the perturbation vector. Based on the feasible region boundary mapping function, the perturbation vector is mapped to the analytical center approximation point of the perturbed feasible region to obtain the candidate attitude compensation gain matrix set corresponding to each perturbation vector. The control module is used to perform numerical processing and vectorization on each matrix in the candidate attitude compensation gain matrix set, and obtain the Gram matrix based on the inner product. The Gram matrix is then decomposed into eigenvalues to obtain the principal eigenvectors as fusion weight vectors. The candidate attitude compensation gain matrix set is combined based on the fusion weight vectors to obtain the global robust gain matrix. The control commands for the chassis actuator are based on the global robust gain matrix, multi-dimensional state data, and the desired target reference value.
