A semi-active suspension control method, device and equipment

Abstract

The application relates to the technical field of intelligent vehicle control, and discloses a semi-active suspension control method, device and equipment, the method comprising the following steps: acquiring current state information of a vehicle suspension system, including sprung mass displacement, sprung mass speed, tire displacement, tire speed and current road excitation; based on a historical road excitation buffer queue, generating a road excitation prediction result in parallel through a first prediction channel and a second prediction channel, and performing weighted fusion on the road excitation prediction result to obtain a fused road excitation prediction sequence; based on the current state information and the fused road excitation prediction sequence, solving a quadratic programming optimization problem through model predictive control to obtain an optimal damping coefficient sequence. The application can improve the road excitation prediction accuracy under complex road conditions, enhance the numerical stability of QP solving, realize intelligent damping adjustment, and thus improve the driving smoothness and control stability of the vehicle.

Description

Technical Field

[0001] This application relates to the field of intelligent vehicle control technology, and in particular to a semi-active suspension control method, device and equipment. Background Technology

[0002] Semi-active suspension systems improve vehicle ride comfort and handling stability by adjusting the damping coefficient in real time. Compared to passive suspensions, they offer better adaptability, while compared to active suspensions, they have lower energy consumption and cost. Model predictive control (MPC) is widely used in semi-active suspension control due to its ability to handle multi-objective optimization and constrained problems. Current technologies for road excitation prediction either rely on external preview sensors to obtain long-term time-domain information or use a single fixed model for prediction. However, a single prediction model struggles to balance prediction accuracy and robustness: when the road surface is stable, the fixed model's accuracy is acceptable; but under complex road conditions such as bumps and speed bumps, the time-varying characteristics of road excitations are significant, and the prediction error of a single model increases sharply. This causes the MPC controller to fail to accurately predict road changes in advance, affecting the control effect. Summary of the Invention

[0003] The purpose of this application is to provide a semi-active suspension control method, device, and equipment to improve the ride comfort and handling stability of a vehicle.

[0004] To achieve the above objectives, the first aspect of this application provides a semi-active suspension control method, comprising the following steps: Obtain the current state vector of the semi-active suspension system, which includes sprung mass displacement, sprung mass velocity, tire displacement, tire velocity, and current road surface excitation value, and update the current road surface excitation value to the historical road surface excitation buffer queue. Based on the historical road surface excitation buffer queue, road surface excitation prediction results are generated in parallel through the first prediction channel and the second prediction channel, and the prediction results of the first prediction channel and the second prediction channel are weighted and fused to generate a road surface excitation prediction sequence in the future prediction time domain. Based on the state vector and the road excitation prediction sequence, with the weighted minimization of sprung acceleration as the optimization objective and the physical limitation of the damping coefficient as the constraint, the optimal damping control quantity sequence is output and applied to the semi-active suspension actuator.

[0005] In some embodiments of this application, the first prediction channel uses an autoregressive model to predict road surface excitation, and the second prediction channel uses a linear extrapolation method to generate compensation prediction. The parameters of the autoregressive model are estimated online using recursive least squares, and the forgetting factor is adaptively adjusted according to the current operating conditions.

[0006] In some embodiments of this application, the parameters of the autoregressive model are estimated online using the recursive least squares method, and the forgetting factor is adaptively adjusted according to the current operating condition confidence indicator; the operating condition confidence indicator is generated based on the amplitude and rate of change of the sprung mass acceleration.

[0007] In some embodiments of this application, the fusion weights of the weighted fusion are determined based on the prediction confidence of the first prediction channel, and the prediction confidence decreases as the prediction error of the first prediction channel increases; the prediction confidence is determined based on the statistical characteristics of the prediction residual sequence of the first prediction channel.

[0008] In some embodiments of this application, an error boundary for road excitation prediction is generated based on the statistical characteristics of the prediction residual sequence of the first prediction channel; based on the error boundary, the state constraints of the quadratic programming optimization problem are adaptively compressed, and the compression coefficients of different state variables are positively correlated with their sensitivity to road excitation errors.

[0009] In some embodiments of this application, the method further includes: adaptively adjusting the historical data window length of the autoregressive model in response to the operating condition confidence indicator, wherein the higher the operating condition level, the shorter the corresponding window length.

[0010] In some embodiments of this application, in response to the operating condition confidence indicator, the weight matrix of the quadratic programming optimization problem is adjusted online. The higher the operating condition level, the greater the weight of the sprung acceleration penalty and the smaller the weight of the control input penalty.

[0011] A second aspect of this application also provides a semi-active suspension control device, comprising: an acquisition module configured to acquire a current state vector of a semi-active suspension system, the state vector including sprung mass displacement, sprung mass velocity, tire displacement, tire velocity, and a current road surface excitation value, and updating the current road surface excitation value to a historical road surface excitation buffer queue; a prediction module configured to generate road surface excitation prediction results in parallel through a first prediction channel and a second prediction channel based on the historical road surface excitation buffer queue, and to perform weighted fusion of the prediction results of the first prediction channel and the second prediction channel to generate a road surface excitation prediction sequence in the future prediction time domain; an optimization module configured to output an optimal damping control quantity sequence based on the state vector and the road surface excitation prediction sequence, with the optimization objective of minimizing the weighted sprung acceleration and the constraint of the physical limitation of the damping coefficient; and an execution module configured to apply the first control quantity of the optimal damping control quantity sequence to the semi-active suspension actuator.

[0012] A third aspect of this application also provides an electronic device including a processor and a memory, wherein computer instructions are stored in the memory, and when the computer instructions are executed by the processor, the electronic device performs the semi-active suspension control method based on the improved MPC in any of the above embodiments.

[0013] A fourth aspect of this application also provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the semi-active suspension control method based on improved MPC in any of the above embodiments.

[0014] The fifth aspect of this application also provides a computer program product tangibly stored on a computer-readable medium and including computer-executable instructions that, when executed, implement the semi-active suspension control method based on improved MPC in any of the above embodiments.

[0015] One or more of the above technical solutions introduce a dual-channel parallel prediction mechanism to predict road excitation separately. The high-quality road excitation prediction sequence after weighted fusion of the prediction results of the two channels is input into the MPC optimization solution stage. This enables the MPC controller to predict the changes in road excitation in the future prediction time domain in advance within the current control cycle, optimize the damping control quantity sequence in advance, and complete the damping adjustment before the road impact reaches the tire, thereby reducing control lag and improving vehicle ride smoothness and passenger comfort. Attached Figure Description

[0016] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0017] Figure 1 This is a schematic diagram of an example environment that can be implemented according to the embodiments of the present invention.

[0018] Figure 2 This is a flowchart of the semi-active suspension control method in an embodiment of the present invention.

[0019] Figure 3 This is a flowchart of road surface excitation prediction based on an AR model. Figure 4 The flowchart shows the damping adjustment strategy based on acceleration. Figure 5 This is a schematic diagram of the acceleration curve of the sprung mass under moderate road conditions. Figure 6 This is a schematic diagram of the vertical displacement curve of the suspension under moderate road conditions. Figure 7This is a schematic diagram of the vertical displacement of the tire under moderate road conditions. Figure 8 This is a program module framework diagram of the semi-active suspension control device in an embodiment of the present invention. Detailed Implementation

[0020] Embodiments of this application will now be described in more detail with reference to the accompanying drawings. While some embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this application. It should be understood that the drawings and embodiments of this application are for illustrative purposes only and are not intended to limit the scope of protection of this application.

[0021] In the description of embodiments of this application, the term "comprising" and similar terms should be understood as open-ended inclusion, i.e., "including but not limited to". The term "based on" should be understood as "at least partially based on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc., may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0022] To facilitate understanding of the technical solution of this application, the core technical terms are defined first.

[0023] Semi-active suspension: refers to a suspension system that improves vehicle ride comfort and handling stability by adjusting the damping coefficient in real time, which is different from passive suspension and active suspension.

[0024] Model predictive control (MPC) predicts the system state in the future time domain by establishing a dynamic model of the system and solving the optimization problem to obtain the optimal control sequence. Only the first control variable is applied to the system, and the process is repeated in the next cycle. It has the ability to handle multi-objective optimization and constraints.

[0025] Autoregressive (AR) model: A time series forecasting model that assumes that the current observation can be represented by a linear combination of observations from several historical times. The model parameters are estimated from historical data and are suitable for predicting road surface excitations with time correlation.

[0026] Recursive Least Squares (RLS): An online parameter estimation algorithm that can quickly update model parameters when new data arrives without storing all historical data. By introducing a forgetting factor mechanism, the model can quickly adapt to time-varying characteristics. It has high computational efficiency and is suitable for real-time control applications.

[0027] Quadratic Programming (QP): A class of optimization problems where the objective function is a quadratic function and the constraints are linear inequalities or equality. Solving the MPC control law is essentially solving a QP problem. Traditional QP solvers are prone to slow convergence or failure when the matrix is ​​ill-conditioned. This application enhances the stability of the solution through regularization and an adaptive step size strategy.

[0028] Sprout mass: refers to the total mass of the vehicle body and its load-bearing components, and its vibration characteristics directly affect passenger comfort.

[0029] Unsprung mass: refers to the mass of wheels, tires, and suspension components, whose vibration characteristics affect road surface adhesion performance.

[0030] As described in the background section, existing technologies mostly employ fixed models or simple autoregressive models for single-channel road excitation prediction, which suffers from problems such as difficulty adapting to time-varying characteristics of complex road conditions and prediction lag. Based on this, this invention proposes a semi-active suspension control scheme. First, the system acquires the current state vector of the semi-active suspension system in real time through onboard sensors. Second, based on a historical road excitation buffer queue, road excitation prediction results are generated in parallel through dual prediction channels and then weighted and fused to generate a road excitation prediction sequence for the future prediction time domain. Subsequently, based on the current state vector and the road excitation prediction sequence, an improved model predictive controller outputs an optimal damping control quantity sequence with the goal of minimizing the weighted sprung acceleration, and the first control quantity is applied to the semi-active suspension actuator. By using dual-channel prediction and weighted fusion, as well as an improved QP solver, the shortcomings of existing technologies are overcome.

[0031] Figure 1 Schematic diagrams are shown illustrating example environments in which various embodiments of this disclosure can be implemented. For example... Figure 1As shown, the example environment includes a vehicle sensor unit 110, an onboard controller 130, and a semi-active suspension actuator 150. The vehicle sensor unit 110, mounted on the vehicle, includes a displacement sensor, a velocity sensor, an acceleration sensor, and a road excitation sensor. It is used to collect real-time status information such as sprung mass displacement, sprung mass velocity, tire displacement, tire velocity, and current road excitation value. The road excitation sensor can be a lidar, camera, or ultrasonic sensor. The onboard controller 130 is an onboard domain controller used to receive the status data collected by the vehicle sensor unit 110, execute the semi-active suspension control method based on the improved MPC of this invention, and output the calculated optimal damping control quantity to the semi-active suspension actuator 150. The semi-active suspension actuator 150 is an adjustable damping shock absorber that adjusts its damping characteristics in real-time according to the damping coefficient control quantity output by the onboard controller 130. The vehicle sensor unit 110 communicates with the vehicle controller 130 via the vehicle CAN bus or vehicle Ethernet, and the vehicle controller 130 communicates with the semi-active suspension actuator 150 via PWM signal, current signal or digital bus.

[0032] The data collection and processing in this application strictly comply with relevant laws and regulations. Data collection, including vehicle status and road surface stimulation, requires authorization and is used solely for semi-active suspension control, not for any other unauthorized purposes. It should be understood that the example environment is described for illustrative purposes only and does not imply any limitation on the scope of this disclosure.

[0033] Figure 2 A schematic flowchart of a semi-active suspension control method 200 according to some embodiments of this application is shown. Example method 200 can be, for example, from... Figure 1 The illustrated vehicle domain controller 130 performs this action. It should be understood that method 200 may also include additional steps not shown, and the scope of this application is not limited in this respect. The following is in conjunction with... Figure 1 Example environment 100 is used to describe method 200 in detail.

[0034] In box 210, the current state vector of the semi-active suspension system is obtained, and the current road surface excitation value is updated to the historical road surface excitation buffer queue.

[0035] The vehicle controller 130 receives real-time data transmitted by the vehicle sensor unit 110 and constructs the state vector of the current time k. and the current road surface excitation value The historical road surface excitation buffer queue is updated using a first-in, first-out (FIFO) mechanism, with the queue length maintained at L steps. The state vector describes the dynamic characteristics of the semi-active suspension system and contains five components, where k represents the current discrete time step. Sprout mass displacement ( Sprout mass velocity (SMS) refers to the vertical displacement of the vehicle body relative to its equilibrium position, reflecting the vertical motion state of the vehicle body, and is collected by displacement sensors installed at the connection points between the vehicle body and the suspension. The vertical displacement of the vehicle body with respect to time is the first derivative of the vehicle body's vertical movement velocity. It can be measured by numerical differentiation of the displacement signal or directly by a speed sensor; tire displacement ( Tire speed () refers to the vertical displacement of the tire's center relative to its equilibrium position, reflecting the tire's vertical motion state, and is collected by a displacement sensor installed at the connection point between the wheel and the suspension; tire speed () This refers to the first derivative of the tire's vertical displacement with respect to time, reflecting the tire's vertical velocity, obtained by numerically differentiating the tire displacement signal; the current road surface excitation ( This refers to the vertical excitation value of road surface unevenness on the tire at the current moment, reflecting the degree of road surface unevenness. It can be collected by lidar, camera, or ultrasonic sensors. All sensors have the same sampling period to ensure data time synchronization, and the data is transmitted to the vehicle controller 130 via the vehicle CAN bus.

[0036] In some embodiments, the sampling period is determined based on control real-time requirements. The prediction time domain is set according to the time-varying characteristics of road surface excitation. and control time domain , The physical constraint range of the damping coefficient is determined based on the physical characteristics of the semi-active suspension actuator. The initial length of the historical road excitation buffer queue is set according to the relevant duration of the road excitation. In this context, the prediction time domain refers to the number of time steps the MPC controller takes to predict future states; the control time domain refers to the number of time steps allowed for adjusting the control quantity in the optimization problem.

[0037] In some embodiments, based on Newton's laws of motion, the continuous-time dynamic equations of the semi-active suspension system are established. The equation of motion for the sprung mass M is: Where k is the suspension stiffness (in N / m) and c is the damping coefficient (control input, in N·s / m). The equation of motion for the unsprung mass m is: ,in Tire stiffness (unit: N / m).

[0038] The continuous-time dynamic equations are discretized using the Euler method, with sampling time... Thus, the discrete state-space model is obtained: ,in This is the damping coefficient (control input). For road surface excitation disturbances, A, B, This represents the discretized system matrices. System matrix A has a dimension of 5×5, and system matrix B has a dimension of 5×1. The dimension is 5×1, and the specific elements are obtained through discretization of the continuous-time dynamic equations. The discretized state-space model provides the predictive basis for subsequent model predictive control. This discrete state-space model describes the evolution of the system state from the current time k to the next time k+1, where the system matrix A describes the natural evolution of the state, the control matrix B describes the influence of the damping coefficient on the state, and the disturbance matrix... This describes the effect of road surface excitation on the state. By recursively applying this model, the future can be obtained. Predicted values ​​for all states within 10 steps.

[0039] The state vector and historical road excitation buffer queue are cached in the internal memory of the on-board controller 130, while the discrete state-space model is stored in the controller's program storage area. This enables synchronous acquisition of multi-source sensor data and construction of state vectors, providing high-quality input data for subsequent dual-channel road excitation prediction and model predictive control. By establishing an accurate discrete state-space model, the prediction accuracy of model predictive control is ensured, improving the overall performance of the semi-active suspension control system. Furthermore, the maintenance of the historical road excitation buffer queue provides sufficient historical data support for road excitation prediction, enhancing the adaptability and robustness of the prediction module.

[0040] In box 220, based on the historical road surface excitation buffer queue, road surface excitation prediction results are generated in parallel through the first prediction channel and the second prediction channel, and then weighted and fused.

[0041] First, the on-board controller 130 generates road excitation prediction results in parallel through the first prediction channel and the second prediction channel based on the historical road excitation buffer queue.

[0042] In some embodiments, such as Figure 3 As shown, the first prediction channel uses an autoregressive (AR) model for road excitation prediction. The model parameters are estimated online using recursive least squares (RLS), and the forgetting factor is adaptively adjusted based on the road condition confidence level. This model is suitable for road excitation sequences with time correlation. For example, the system updates the AR model coefficients online using recursive least squares (RLS) based on data from the historical road excitation buffer queue. The AR model is in the following form:

[0043] Where a, b, c, and d are model coefficients. For the noise term (assumed to be zero-mean Gaussian white noise), the first derivative of the road surface excitation is... and second derivative The data was obtained by numerical differentiation of historical road surface excitation sequences, and a central difference scheme was used to improve accuracy.

[0044] The system updates the AR model coefficient vector online using recursive least squares (RLS) with a forgetting factor, based on data from the historical road surface excitation buffer queue. The eigenvectors consist of the current historical road surface excitation values ​​and their first and second derivatives. The initial covariance matrix is ​​set to a large positive definite matrix to ensure the initial convergence speed, and the initial parameter vectors are set empirically.

[0045] Among them, the forgetting factor The model dynamically adjusts based on the working condition confidence level indicator to adapt to the time-varying characteristics of different road conditions. The working condition confidence level indicator refers to a multi-level vibration intensity rating (including severe, moderate, slight, stable, and normal levels) generated based on the amplitude and rate of change of sprung mass acceleration. This rating characterizes the vibration intensity of the current road surface excitation and drives the adaptive adjustment of the prediction and control modules. For example, under severe vibration conditions, the forgetting factor is set to a smaller value, allowing the model to quickly forget historical data and adapt to new road surface characteristics. As the vibration level decreases, the forgetting factor gradually increases. Under stable or normal conditions, the forgetting factor is set to a larger value, fully utilizing historical data to improve prediction accuracy. The forgetting factor ranges from (0, 1), with a smaller value indicating a faster rate of forgetting of historical data. Through adaptive adjustment of the forgetting factor, the AR model can quickly respond to road surface changes under severe vibration conditions and fully utilize historical information to improve prediction stability under stable conditions.

[0046] Based on the updated AR model coefficients Recursive prediction of the future Step-by-step road surface excitation sequence Each step of the prediction is calculated by substituting the previous prediction value and the corresponding first and second derivative terms into the AR model. The initial required historical values ​​are taken from the historical road surface excitation buffer queue.

[0047] For example, the process of generating the operating condition confidence label is as follows: The on-board controller 130 reads the absolute value of the sprung mass acceleration at the current moment. and its rate of change acceleration of the sprung mass By measuring the velocity of the spring mass The rate of change of acceleration is obtained by numerical differentiation. The acceleration sequence of the spring mass is obtained by numerical differentiation. The system operates according to a preset first threshold. Second threshold Third threshold ( The current vibration condition is judged into five levels, and the condition confidence index is generated: (1) Severe: or Exceeding the first threshold (1) indicates that the current road is in a bumpy section or speed bump, etc., and is subject to severe impact conditions; (2) medium: Exceeding the second threshold But it did not exceed the first threshold. ,and Not exceeding the first threshold (3) Slight: This indicates that the current vibration condition is moderate; Exceeding the third threshold But it did not exceed the second threshold. This indicates that the current operating condition is under slight vibration; (4) Stable: and All are extremely small (none exceed the third threshold) (5) Normal: Other situations, that is, the current vibration state is within the normal range. The first threshold Second threshold Third threshold The values ​​are calibrated based on the vehicle's sprung mass characteristics and passenger comfort requirements, with the specific values ​​determined by the actual vehicle model parameters.

[0048] In some embodiments, the second prediction channel generates compensation predictions using a linear extrapolation method. Extrapolation is performed based on the current pavement excitation value and its first and second derivatives, assuming the pavement condition maintains its current trend to ensure the continuity of the prediction process. The prediction sequence generated by the second prediction channel is as follows: .

[0049] Then, the prediction results of the first and second prediction channels are weighted and fused to generate a road surface excitation prediction sequence in the future prediction time domain. Specifically, for each prediction step in the future time domain, the prediction values ​​of the two channels at the same step are weighted and calculated using corresponding weights. In some embodiments, the fusion weights of the weighted fusion are determined based on the prediction confidence of the first prediction channel, and the prediction confidence decreases as the prediction error of the first prediction channel increases.

[0050] For example, prediction confidence refers to the reliability of the prediction result from the first prediction channel, calculated based on the statistical characteristics of the prediction residual sequence, and decreases as the prediction error increases. The prediction residual is defined as the difference between the actual road surface excitation value at the current moment and the predicted road surface excitation value at the previous moment, i.e. Based on predicted residual sequences The root mean square error (RMSE) is used to calculate the prediction confidence of the first prediction channel. The prediction confidence level decreases monotonically with increasing RMSE, and its value ranges from [0, 1]; the weight of the second prediction channel... .

[0051] To predict confidence and Using the weights, the prediction sequences of the first and second prediction channels are weighted and fused to obtain the fused road excitation prediction sequence. When the prediction accuracy of the first prediction channel is high (RMSE is small), (Approaching 1), the fusion result mainly relies on AR model prediction; when the prediction accuracy of the first prediction channel is low (RMSE is large), (Approaching 0), the fusion result mainly relies on linear extrapolation prediction to ensure the robustness of the prediction result.

[0052] Therefore, by using a dual-channel parallel prediction and confidence-weighted fusion mechanism, combined with online estimation of the AR model and linear extrapolation compensation, the time-varying characteristics of complex road conditions can be effectively captured, and the prediction error is significantly reduced.

[0053] In some embodiments, when historical data is insufficient (e.g., in the early stages of system startup, the length of the historical road surface excitation buffer queue is less than 3 steps) or the confidence level of the first prediction channel is low, a linear extrapolation formula is used to predict the future. The surface excitation is applied. The linear extrapolation formula is:

[0054] in and These are the first and second derivatives of the road surface excitation at the current moment, obtained by numerical differentiation of the three most recent historical road surface excitation values. Based on the assumption that the current road surface condition maintains its current trend, the continuity of the prediction process is ensured, avoiding control interruptions due to missing data or failure of the first prediction channel.

[0055] In some embodiments, the operating condition confidence level is used as an adjustment and arbitration signal for the historical data window length of the autoregressive model and the weights of the linear extrapolation channels during dual-channel fusion. When the operating condition confidence level is severe, the system shortens the historical data window length used by the autoregressive model and increases the fusion weights corresponding to the linear extrapolation of the second prediction channel, thereby weakening the adverse interference of long-term stable historical data on the prediction results of bumpy and sudden road conditions. When the operating condition confidence level is stable or normal, the system restores the historical data window length and the initial dual-channel fusion weights to the preset standard configuration.

[0056] In box 230, a standard quadratic programming (QP) problem is constructed and solved with the objective function of minimizing the weighted acceleration of the spring and the physical constraint of the damping coefficient as the inequality constraint, so as to obtain the optimal damping coefficient sequence in the future control time domain.

[0057] In some embodiments, to solve the QP optimization problem described above, it is necessary to first construct a prediction matrix based on the discrete state-space model established in block 210, convert the dynamic equations of the semi-active suspension system into matrix form, and establish a mapping relationship between the control input and the future state. The prediction matrix includes three types: state transition matrix. Describe the influence of the current state on future states; control input influence matrix Describe the impact of the control sequence on future states; perturbation effect matrix This describes the impact of the road surface excitation prediction sequence on future states. These three types of prediction matrices are obtained through a recursive expansion of the discrete state-space model, and their specific elements are derived from the system matrices A, B, and C. The power and combination of the terms are calculated using a standard matrix recursion algorithm.

[0058] In some embodiments, the objective function and constraints for MPC are constructed based on the prediction matrix. The design principle of the objective function is to minimize the weighted average of sprung acceleration as the optimization objective, while penalizing excessive control inputs, balancing vehicle ride comfort and energy consumption. The formula is: The objective function formula is:

[0059] Where C is the output matrix (extracting the sprung velocity component and obtaining the sprung acceleration through numerical differentiation), Q is the state weight matrix, and R is the control weight matrix. The state weight matrix Q is a diagonal matrix, with diagonal elements corresponding to the weights of sprung displacement, sprung velocity, tire displacement, tire velocity, and road excitation, respectively. The weight corresponding to sprung velocity is significantly higher than that of other state weights to focus on the optimization objective of sprung mass acceleration. The control weight matrix R is set to a small value to balance comfort and energy consumption.

[0060] Specifically, the objective function for the future The spring mass accelerations at each step within the time limit are weighted and summed, with the weighting coefficients determined by the state weight matrix Q. Simultaneously, the control time domain is... The control input (damping coefficient) at each step is penalized, and the penalty coefficient is determined by the control weight matrix R. The weight of sprung speed is significantly higher than that of other state weights because the rate of change of sprung speed (i.e., sprung acceleration) is a key indicator for evaluating ride comfort. The control weight R is set to a small value to ensure that the optimization process prioritizes ride comfort.

[0061] In some embodiments, the improved QP solver employs two core technologies: regularization and an adaptive step-size iteration strategy, to enhance the numerical stability and convergence speed of the solution. The purpose of regularization is to prevent solution failure when the Hessian matrix H approaches singularity. Regularization involves adding small regularization coefficients to the diagonal of the Hessian matrix. This forms a regularized Hessian matrix:

[0062] in It is a small positive number (ensuring the matrix is ​​strictly positive definite). for The identity matrix. This operation ensures that the Hessian matrix is ​​strictly positive definite, thus guaranteeing the convexity and solvability of the QP problem. The adaptive step-size iteration strategy aims to accelerate the solution speed as much as possible while ensuring convergence stability.

[0063] Traditional fixed-step-size iterative methods may encounter problems when the curvature of the objective function changes significantly, such as oscillations due to excessively large step sizes or slow convergence due to excessively small step sizes. The adaptive step-size strategy dynamically adjusts the step size based on the descent of the objective function: when the objective function decreases, it indicates that the current step size is appropriate and can be more aggressive, so the step size is appropriately increased; when the objective function increases, it indicates that the current step size is too large and may overshoot the optimal point, so the step size is decreased. The damping coefficient is initialized to the intermediate value of the physical constraints. In each iteration, the gradient is calculated and the control sequence is updated along the negative gradient direction. After the update, the control sequence is ensured to be within the physical constraints (if it exceeds, it is projected onto the constraint boundary). Iteration stops when the difference between the control sequences of two consecutive iterations is less than the convergence threshold or the number of iterations exceeds the upper limit, and the optimal solution is output. Through the two improvements mentioned above, the numerical stability of the QP solver is significantly enhanced.

[0064] The improved QP solver employs regularization and an adaptive step-size iteration strategy to enhance the numerical stability of the solution. Regularization: A small regularization coefficient is added to the Hessian matrix to form a regularized Hessian matrix. ( It is a small positive number. for The identity matrix is ​​used to prevent the solution from failing when the Hessian matrix is ​​close to singular. The adaptive step-size iteration strategy includes the following four steps: (1) Initialization: Initialize the damping coefficient sequence to the intermediate value of the physical constraint. (2) Gradient calculation: (3) Iterative update: ,in The step size is adaptively adjusted based on the decrease of the objective function. If the objective function decreases, the step size is increased appropriately; if the objective function increases, the step size is decreased. After the update, ensure... Within physical constraints (if exceeding, project to the constraint boundary); (4) Convergence judgment: Stop when the difference between the control sequences of two consecutive iterations is less than the convergence threshold or the number of iterations exceeds the upper limit, and output the optimal solution. .

[0065] Output optimal damping control sequence This sequence contains the optimal damping coefficients for 8 steps within the future control time domain. Based on the rolling optimization principle of MPC, only the first control variable is extracted as the basic optimal damping coefficient and passed to block 240 for multi-level dynamic adjustment. The control variables are transmitted to the semi-active suspension actuator 150 through the output interface of the onboard controller 130. The remaining control variables in the sequence are used as initial guesses for the rolling optimization of the next control cycle and stored in the internal memory of the onboard controller 130 to accelerate the solution speed of the next cycle.

[0066] The final damping coefficient is strictly limited to the range of physical constraints. Within the range, if the damping coefficient exceeds the limit, it is projected to the nearest constraint boundary. This operation ensures that the damping coefficient does not exceed the physical operating range of the adjustable damper, avoiding damage to the damping valve or suspension failure due to over-adjustment.

[0067] In some embodiments, the weight matrices Q and R are adjusted online via a second feedback channel using operating condition confidence indicators generated from block 220: under severe vibration conditions, the sprung speed penalty weight is increased and the control input penalty weight is decreased, enabling the controller to more actively suppress vibration; under moderate operating conditions, the sprung speed penalty weight is moderately increased and the control input penalty weight R is moderately decreased, achieving a moderately comfortable balance between comfort and energy consumption; under stable or normal operating conditions, the default weights are restored, prioritizing both energy consumption and handling stability. These adjustment rules ensure that higher operating condition levels correspond to larger sprung acceleration penalty weights and smaller control input penalty weights, forming a progressive adaptive weight adjustment mechanism.

[0068] In some embodiments, the system further generates the error boundary for road excitation prediction based on the statistical characteristics of the predicted residual sequence. Adaptive tightening of the MPC state constraints is applied: a larger prediction error boundary results in a smaller allowable range for the state constraints, reserving a greater safety margin for suspension motion; a smaller prediction error boundary results in a larger allowable range for the state constraints, fully utilizing suspension travel to improve comfort. The error boundary is calculated as follows: ,in To predict the standard deviation of the residual series, For example, the confidence interval coefficients That is, using twice the standard deviation as the error boundary, the error boundary represents the uncertainty range of the road excitation prediction. The smaller the error boundary, the more accurate and stable the prediction is, and the higher the credibility. The larger the error boundary, the more unstable the prediction result is, and the lower the credibility. Based on this, the above error boundary can be used for subsequent adaptive tightening of MPC state constraints.

[0069] For example, assume the original state constraints are ( ),in and These represent the upper and lower limits of the physical constraints for each component of the state vector. The constraint range for each state variable is adaptively tightened according to the error boundary, and the tightening coefficient of different state variables is positively correlated with their sensitivity to road excitation prediction errors.

[0070]

[0071] in, and The upper and lower limits of physical constraints after compressing each component of the state vector. This refers to the compression coefficient. For example, the compression coefficient can be calibrated through multi-road condition simulation. By solving for the steady-state sensitivity of each state relative to the road excitation disturbance, the sensitivity is normalized and mapped to obtain the corresponding compression coefficient. The more sensitive the state is to the prediction error, the larger the compression coefficient, and the stronger the corresponding constraint compression. The road excitation itself does not participate in constraint adjustment, and its compression coefficient is set to 0. The larger the prediction error boundary, the smaller the allowable constraint range, reserving a larger safety margin for suspension movement; the smaller the prediction error boundary, the larger the allowable constraint range, fully utilizing the suspension travel to improve comfort. Through adaptive compression of state constraints, the constraint range is automatically tightened when the uncertainty of road excitation prediction increases, improving system robustness and preventing control failure due to state out-of-bounds errors.

[0072] Therefore, by generating error boundaries based on the statistical characteristics of the prediction residuals and embedding them into the adaptive tightening mechanism of the MPC state constraints, the constraints are automatically tightened when the prediction uncertainty increases, which helps to improve the robustness of the system.

[0073] In some embodiments, when the error boundary exceeds a preset threshold, the system triggers a conservative control strategy to address the control risks caused by prediction inaccuracies. The conservative control strategy includes the following three measures: First, freezing AR model parameter updates, stopping RLS recursive updates, and maintaining the most recently valid model coefficients. First, keep the model coefficients unchanged to avoid oscillations when predictions are inaccurate. Second, increase the constraint tightening to further tighten the state constraints and improve control conservatism. Third, fix the damping coefficient and force it to be set to a relatively large value within the physical constraint range, suspend MPC optimization, and ensure that the vehicle can still maintain basic vibration reduction performance when predictions are inaccurate.

[0074] In some embodiments, when the error boundary is less than a preset proportion of the instability threshold for several consecutive control cycles, the conservative control strategy is lifted, and a gradual recovery mechanism is adopted to gradually restore the normal control mode. The specific recovery steps are as follows: first, AR model parameter updates are resumed, and RLS recursive updates are restarted; then, the constraint compression coefficient is gradually reduced, decreasing by a certain step size each control cycle until it returns to the normal value; finally, MPC optimization is resumed, and the damping coefficient optimization calculation is restarted. The gradual recovery mechanism avoids abrupt changes in the control mode, ensuring a smooth transition in the control process.

[0075] Therefore, by using a conservative control strategy trigger mechanism, the system can switch to conservative control mode in a timely manner when predictions are inaccurate, thereby avoiding control failure and ensuring vehicle safety.

[0076] Understandably, if a conservative control strategy is triggered, i.e., the road surface excitation prediction is inaccurate, the system will directly output a fixed damping coefficient and repeat the process. This process generates a control sequence to ensure that the vehicle maintains basic damping performance even when predictions are inaccurate.

[0077] In some embodiments, such as Figure 4 As shown, based on the working condition confidence level, the optimal damping coefficient of the foundation is also determined. The final damping coefficient is obtained by performing multi-level dynamic adjustments. (1) Severe vibration condition: Significantly increase the damping coefficient, Adjust to the first adjustment value to quickly suppress severe vibrations and prevent vehicle pitching or nodding; (2) Medium vibration conditions: moderately increase the damping coefficient to Adjust to the second adjustment value (the second adjustment value is less than the first adjustment value) to balance comfort and handling; (3) Slight vibration conditions: reduce the damping coefficient, Adjust to the third adjustment value to improve driving comfort; (4) Stable operating condition: further reduce the damping coefficient, and Adjusted to the fourth adjustment value (the fourth adjustment value is less than the third adjustment value) to enhance tire-ground contact and reduce energy consumption; (5) Normal working condition: directly use the QP solution result of box 230, that is This ensures the economy and effectiveness of control.

[0078] By using the operating condition confidence label, the system can correct the front-end prediction strategy and optimization target based on the real-time perceived vibration operating conditions, correct the constraint boundary of the state vector, and dynamically adjust the damping coefficient in multiple levels. Without external pre-aiming sensors, the system achieves self-predictive, self-optimizing, and self-robust semi-active suspension control, and keeps prediction and execution synchronized when road conditions change abruptly.

[0079] Corresponding to the above-described semi-active suspension control method embodiment, the present invention also provides an amphibious vehicle searchlight guidance navigation control device 300, such as... Figure 8 As shown, the device includes: an acquisition module 310, configured to acquire the current state vector of the semi-active suspension system, the state vector including sprung mass displacement, sprung mass velocity, tire displacement, tire velocity, and current road surface excitation value, and update the current road surface excitation value to a historical road surface excitation buffer queue; a prediction module 320, configured to generate road surface excitation prediction results in parallel through a first prediction channel and a second prediction channel based on the historical road surface excitation buffer queue, and to perform weighted fusion of the prediction results of the first prediction channel and the second prediction channel to generate a road surface excitation prediction sequence in the future prediction time domain; an optimization module 330, used to output an optimal damping control quantity sequence based on the state vector and the road surface excitation prediction sequence, with the optimization objective of minimizing the weighted sprung acceleration and the constraint of the physical limitation of the damping coefficient; and an execution module 340, used to apply the first control quantity of the optimal damping control quantity sequence to the semi-active suspension actuator.

[0080] This invention also provides an electronic device including a computing unit capable of performing various appropriate actions and processes based on computer program instructions stored in random access memory (RAM) and / or read-only memory (ROM) or loaded from a storage unit into RAM and / or ROM. Various programs and data required for device operation may also be stored in the RAM and / or ROM. The computing unit and the RAM and / or ROM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0081] Multiple components within the device are connected to I / O interfaces, including: input units such as a steering wheel angle sensor, pedal position sensor, and ECG acquisition module; output units such as a central control display, speaker, and hazard warning lights; storage units such as a solid-state drive and embedded memory; and communication units such as an in-vehicle Ethernet controller, Bluetooth module, and V2X communication module. The communication units allow the device to exchange information / data with other devices via computer networks such as the Internet and / or various telecommunications networks.

[0082] A computing unit can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of computing units include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit performs the various methods and processes described above, such as method 200. For example, in some embodiments, method 200 may be implemented as a computer software program tangibly contained in a machine-readable medium. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via RAM and / or ROM and / or a communication unit. When the computer program is loaded into RAM and / or ROM and executed by the computing unit, one or more steps of method 200 described above may be performed. Alternatively, in other embodiments, the computing unit may be configured to perform method 200 by any other suitable means (e.g., by means of firmware).

[0083] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When these computer program instructions are loaded and executed on a server or terminal, they generate, in whole or in part, the processes or functions described in the embodiments of this application. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic cable, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to the server or terminal, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, and magnetic tape), an optical medium (e.g., digital video disk (DVD), etc.), or a semiconductor medium (e.g., solid-state drive).

[0084] Furthermore, although the operations are described in a specific order, this should be understood as requiring that such operations be performed in the specific order shown or in sequential order, or requiring that all illustrated operations be performed to achieve the desired result. In certain environments, multitasking and parallel processing may be advantageous. Similarly, although several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of the invention. Certain features described in the context of individual embodiments may also be implemented in combination in a single implementation. Conversely, various features described in the context of a single implementation may also be implemented individually or in any suitable sub-combination in multiple implementations.

[0085] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

Claims

1. A semi-active suspension control method, characterized in that, Includes the following steps: Obtain the current state vector of the semi-active suspension system, which includes sprung mass displacement, sprung mass velocity, tire displacement, tire velocity, and current road surface excitation value, and update the current road surface excitation value to the historical road surface excitation buffer queue. Based on the historical road surface excitation buffer queue, road surface excitation prediction results are generated in parallel through the first prediction channel and the second prediction channel, and the prediction results of the first prediction channel and the second prediction channel are weighted and fused to generate a road surface excitation prediction sequence in the future prediction time domain. Based on the state vector and the road excitation prediction sequence, with the weighted minimization of sprung acceleration as the optimization objective and the physical limitation of the damping coefficient as the constraint, the optimal damping control quantity sequence is output and applied to the semi-active suspension actuator.

2. The method according to claim 1, characterized in that, The first prediction channel uses an autoregressive model to predict road surface excitation, while the second prediction channel uses a linear extrapolation method to generate compensation prediction.

3. The method according to claim 2, characterized in that, The parameters of the autoregressive model are estimated online using the recursive least squares method, and the forgetting factor is adaptively adjusted based on the current operating condition confidence indicator; the operating condition confidence indicator is generated based on the amplitude and rate of change of the sprung mass acceleration.

4. The method according to claim 1, characterized in that, The fusion weights of the weighted fusion are determined based on the prediction confidence of the first prediction channel, which decreases as the prediction error of the first prediction channel increases; the prediction confidence is determined based on the statistical characteristics of the prediction residual sequence of the first prediction channel.

5. The method according to claim 4, characterized in that, Furthermore, an error boundary for road excitation prediction is generated based on the statistical characteristics of the prediction residual sequence of the first prediction channel; based on the error boundary, the state constraints of the quadratic programming optimization problem are adaptively compressed, and the compression coefficients of different state variables are positively correlated with their sensitivity to road excitation errors.

6. The method according to claim 3, characterized in that, Also includes: In response to the operating condition confidence indicator, the historical data window length of the autoregressive model is adaptively adjusted, with a shorter window length corresponding to a higher operating condition level.

7. The method according to claim 3, characterized in that, In response to the operating condition confidence indicator, the weight matrix of the quadratic programming optimization problem is adjusted online. The higher the operating condition level, the greater the weight of the sprung acceleration penalty and the smaller the weight of the control input penalty.

8. A semi-active suspension control device, characterized in that, include: The acquisition module is configured to acquire the current state vector of the semi-active suspension system, the state vector including sprung mass displacement, sprung mass velocity, tire displacement, tire velocity and current road surface excitation value, and update the current road surface excitation value to the historical road surface excitation buffer queue; The prediction module is configured to generate road excitation prediction results in parallel through a first prediction channel and a second prediction channel based on the historical road excitation buffer queue, and to perform weighted fusion of the prediction results of the first prediction channel and the second prediction channel to generate a road excitation prediction sequence in the future prediction time domain. The optimization module is used to output the optimal damping control quantity sequence based on the state vector and the road surface excitation prediction sequence, with the optimization objective of minimizing the weighted sprung acceleration and the physical constraint of the damping coefficient. The execution module is used to apply the first control quantity of the optimal damping control quantity sequence to the semi-active suspension actuator.

9. An electronic device, characterized in that, include: processor; A memory for storing executable instructions of the processor; wherein the processor is configured to perform the method of claim 1 by executing the executable instructions.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of claim 1.

Images