Dynamic force feedback closed-loop control method and system in simulated driving

Abstract

The invention provides a dynamic force feedback closed-loop control method and system in simulated driving, and the method comprises the steps: obtaining vehicle model parameters, a steering wheel angle measurement value, and virtual sensor data, such as vehicle speed and transverse acceleration, output by a vehicle solver, and constructing a state space submodel suitable for low-speed, medium-speed and high-speed working conditions; a volume Kalman filter is configured for each sub-model; calculating the confidence coefficient weight of each sub-model in combination with the vehicle speed and the transverse acceleration, and further performing weighted fusion to obtain a global optimal state estimation and equivalent linear system model; high-frequency energy of a steering wheel is extracted through short-time Fourier transform to serve as driver input characteristics, a cost function weight matrix of a linear quadratic regulator is determined through a lookup table, an algebraic Riccati equation is solved to obtain an optimal feedback gain, and then the target output torque of a force feedback steering wheel motor is calculated.

Description

Technical Field

[0001] The present application relates to the field of control, and in particular to a dynamic force feedback closed-loop control method and system in simulated driving. Background Art

[0002] The force feedback steering wheel in a car driving simulator accurately transmits the steering system torque to the driver in real time based on the driving state of the virtual vehicle, thereby simulating real-world road feel and steering feel. The quality of the force feedback control method determines the realism and effectiveness of the simulated driving. There are various ways to implement force feedback. One approach is based on empirical formulas or lookup tables. These methods directly map a few parameters, such as vehicle speed and steering wheel angle, to output torque using a pre-set mapping relationship. This approach is simple to implement and requires minimal computation, but it is overly simplistic and cannot accurately reproduce the complex dynamic characteristics of the vehicle under different operating conditions. In particular, when the vehicle enters a nonlinear region, the feedback torque becomes significantly disconnected from the actual dynamic response of the virtual vehicle, resulting in a distorted driving experience and a failure to provide the driver with accurate vehicle status information. Another approach incorporates a vehicle dynamics model and combines it with modern control theory to generate the feedback torque, such as by using PID control or a linear quadratic regulator (LQR). While theoretically more advanced, these methods are typically nonlinear and time-varying systems, with dynamic characteristics that vary significantly under low, medium, and high speed conditions. Using a single linear model for full-condition control will inevitably lead to model mismatch problems and reduce control accuracy. In the state estimation link, in order to obtain the state variables required by the controller, such as the center of mass sideslip angle, observers such as Kalman filtering are generally used. However, traditional filtering algorithms mostly use a fixed process and measurement noise covariance matrix, which cannot adapt to changes in operating conditions such as tire state changes, resulting in low state estimation accuracy. Moreover, the weighting matrix in the performance indicators of existing optimal control methods is usually a fixed value set offline, and cannot be adjusted online according to the vehicle state and the driver's driving intention, making the control strategy relatively rigid and difficult to always provide better force feedback in changing driving scenarios. Summary of the Invention

[0003] In a first aspect of the present invention, a dynamic force feedback closed-loop control method in simulated driving is provided, comprising the following steps: Obtaining vehicle model parameters, real-time steering wheel angle measurements, and virtual sensor data output by a vehicle solver, including vehicle speed and lateral acceleration; establishing state-space submodels corresponding to low-speed, medium-speed, and high-speed driving conditions based on the vehicle model parameters, configuring a corresponding cubic Kalman filter for each submodel, and determining the process and measurement noise covariance matrix of the cubic Kalman filter using tire state and steering wheel angular velocity; Calculating confidence weights for each sub-model based on the vehicle speed and lateral acceleration in the virtual sensor data, and using the confidence weights to perform weighted fusion on the state estimation vectors output in parallel by each filter and the system matrix of each sub-model to obtain a global optimal state estimation vector and an equivalent linear system model; Performing a short-time Fourier transform on the real-time steering wheel angle measurement value, and using the high-frequency energy ratio as a driver input characteristic parameter; constructing a linear quadratic regulator, and determining a state weighting matrix and a control weighting matrix in a cost function of the linear quadratic regulator based on a three-dimensional lookup table, wherein the inputs of the lookup table are the vehicle speed and center of mass sideslip angle components in the global optimal state estimation vector, and the driver input characteristic parameter; The algebraic Riccati equation is solved based on the equivalent linear system model, the state weighting matrix and the control weighting matrix to obtain the optimal feedback gain matrix, and the gain matrix is ​​multiplied by the global optimal state estimation vector to obtain the target output torque of the force feedback steering wheel actuator motor.

[0004] In a second aspect of the present invention, a dynamic force feedback closed-loop control system for simulated driving is proposed, comprising the following modules: A configuration module is configured to obtain vehicle model parameters, real-time steering wheel angle measurements, and virtual sensor data output by a vehicle solver, including vehicle speed and lateral acceleration; based on the vehicle model parameters, establish state-space submodels corresponding to low-speed, medium-speed, and high-speed driving conditions, configure a corresponding cubic Kalman filter for each submodel, and determine the process and measurement noise covariance matrix of the cubic Kalman filter using tire state and steering wheel angular velocity; a model building module for calculating confidence weights for each sub-model based on the vehicle speed and lateral acceleration in the virtual sensor data, and using the confidence weights to perform weighted fusion on the state estimation vectors output in parallel by each filter and the system matrix of each sub-model to obtain a global optimal state estimation vector and an equivalent linear system model for controller design; a transformation module for performing a short-time Fourier transform on the real-time steering wheel angle measurement value, using the high-frequency energy ratio as a driver input characteristic parameter; constructing a linear quadratic regulator, and determining a state weighting matrix and a control weighting matrix in a cost function of the linear quadratic regulator based on a three-dimensional lookup table, wherein the inputs of the lookup table are the vehicle speed and center of mass sideslip angle components in the global optimal state estimation vector, and the driver input characteristic parameter; A control module is used to solve the algebraic Riccati equation based on the equivalent linear system model and the state weighting matrix and the control weighting matrix to obtain an optimal feedback gain matrix, and multiply the gain matrix with the global optimal state estimation vector to obtain a target output torque of the force feedback steering wheel actuator motor.

[0005] The present invention establishes and weightedly fuses multiple vehicle state space sub-models covering low, medium and high speeds, which can more effectively cover the equivalent system model of the vehicle's nonlinear characteristics under the entire operating condition range; a volumetric Kalman filter with noise parameters that vary with the operating condition is configured for each sub-model, thereby improving the estimation accuracy of key state variables such as the center of mass sideslip angle; the extracted driver input characteristic parameters are used together with the vehicle state as the query basis for the weighted matrix of the linear quadratic regulator cost function, so that the solution of the optimal control law can take into account the vehicle's stability and the driver's manipulation intention. The generated feedback torque not only matches the motion state of the virtual vehicle, but also responds to differences in driving style, thereby improving the road feel authenticity of the simulated driving. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Figure 1 Flowchart for the calculation of process noise covariance matrix and measurement noise covariance matrix; Figure 2 Schematic diagram of model confidence calculation; Figure 3 Schematic diagram of steering wheel angle signal and corresponding spectrum and frequency band; Figure 4 Schematic diagram for weighting matrix determination. DETAILED DESCRIPTION

[0007] In order to make the purpose, technical solutions and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with the specific embodiments of this application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application. It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0008] In a first embodiment of the present invention, a dynamic force feedback closed-loop control method in simulated driving is provided, such as Figure 1 As shown, the following steps are included: S1, obtaining vehicle model parameters, real-time steering wheel angle measurements, and virtual sensor data including vehicle speed and lateral acceleration output by a vehicle solver; based on the vehicle model parameters, establishing state-space submodels corresponding to low-speed, medium-speed, and high-speed driving conditions, respectively, and configuring a corresponding cubic Kalman filter for each submodel; and determining the process and measurement noise covariance matrix of the cubic Kalman filter using tire state and steering wheel angular velocity; Different vehicles have different driving experiences. For example, the actual driving experience differs between SUVs and compact cars, and these differences need to be reflected in driving simulations. Vehicle model parameters such as vehicle mass, moment of inertia, wheelbase, and front and rear wheel cornering stiffness are pre-set by consulting the vehicle design manual. Real-time steering wheel angle measurements are obtained by sampling a photoelectric encoder mounted on the simulator's steering shaft at a frequency of 1000 Hz. Virtual sensor data is calculated in real time using vehicle dynamics simulation software such as CarSim or Adams, outputting the current longitudinal vehicle speed and lateral acceleration at the center of mass at each simulation step.

[0009] After the vehicle model is established, a two-degree-of-freedom vehicle dynamics bicycle model is used as the basis, and three state space sub-models are obtained by linearizing at different vehicle speeds and tire side slip angles. For example, the low-speed model corresponds to 0 to 40 kilometers per hour, the medium-speed model corresponds to 40 to 120 kilometers per hour and a smaller tire side slip angle, and the high-speed nonlinear model corresponds to more than 120 kilometers per hour or a larger tire side slip angle. An independent volumetric Kalman filter is configured for each sub-model for state estimation. In one embodiment, the adjustment of the process noise covariance matrix Q and the measurement noise covariance matrix R is achieved through a fuzzy logic inference system. The fuzzy logic inference system uses the steering wheel angular velocity and the estimated tire side slip degree as input. When the steering wheel angular velocity is large or the tire enters the nonlinear region, the value of the Q matrix is ​​increased and the credibility of the model is reduced, and vice versa. Figure 1 In one embodiment, the linearization described above is specifically that low speed and medium speed are linear regions, and high speed is nonlinear. In another embodiment, low speed and high speed are nonlinear, and medium speed is linear. Alternatively, both speeds may be linearized, and the present invention does not make specific limitations on this.

[0010] S2, calculating the confidence weight of each sub-model based on the vehicle speed and lateral acceleration in the virtual sensor data, and using the confidence weight to perform weighted fusion on the state estimation vectors output in parallel by each filter and the system matrix of each sub-model to obtain the global optimal state estimation vector and the equivalent linear system model for controller design; The confidence weight of each sub-model is calculated using the Gaussian membership function. For example, three Gaussian functions are established for vehicle speed, with 20, 80, and 140 kilometers per hour as the center, respectively. Figure 2As shown in the figure, the weight of each model in the vehicle speed dimension is calculated according to the current vehicle speed. Similarly, a membership function is established for the lateral acceleration. The weights in the two dimensions are multiplied and normalized to obtain the final weight coefficients w1, w2, and w3. The global optimal state estimation vector is obtained by the formula Calculated, where x1, x2, x3 are the outputs of the Kalman filters of each volume. The system matrix of the equivalent linear system model is preferably obtained by as well as Obtained by weighted average.

[0011] S3, performing a short-time Fourier transform on the real-time steering wheel angle measurement value, and using the high-frequency energy ratio as the driver input characteristic parameter; constructing a linear quadratic regulator, and determining a state weighting matrix and a control weighting matrix in a cost function of the linear quadratic regulator based on a three-dimensional lookup table, wherein the inputs of the lookup table are the vehicle speed and center of mass sideslip angle components in the global optimal state estimation vector, and the driver input characteristic parameter; The steering wheel angle signal sequence within the past second is intercepted, and a short-time Fourier transform is performed after applying a window function to obtain the spectrum distribution of the signal. The frequency band from 0 to 10 Hz is divided into a low frequency band from 0 to 2 Hz and a high frequency band from 2 to 10 Hz, as shown in the following example: Figure 3 As shown, the energy within the two frequency bands, or the sum of the squares of the spectrum amplitudes, is calculated. The high-frequency energy is divided by the total energy. The resulting high-frequency energy ratio is the driver input characteristic parameter, which ranges from 0 to 1. A larger value indicates more aggressive driver control.

[0012] State weighting matrix in the linear quadratic regulator LQR cost function J and control weight matrix The element values ​​of are obtained by online interpolation from a pre-calibrated 3D lookup table. The three input dimensions of the lookup table are vehicle speed, center of mass side slip angle, and driver input characteristic parameters, such as Figure 4 For example, when the vehicle speed is high and the sideslip angle is large, the lookup table will output a state with a large penalty term for the sideslip angle. matrix to prioritize vehicle stability; when the driver inputs a larger characteristic parameter, the lookup table will output a smaller one matrix to reduce the constraints on the amount of control, allowing the system to respond faster.

[0013] S4, solving the algebraic Riccati equation based on the equivalent linear system model and the state weighting matrix and the control weighting matrix to obtain the optimal feedback gain matrix, and multiplying the gain matrix with the global optimal state estimation vector to obtain the target output torque of the force feedback steering wheel actuator motor.

[0014] In each control cycle, the equivalent system matrix obtained by weighted fusion is 、 The weighted matrix obtained by table lookup 、 Substituting into the algebraic Riccati equation , use numerical iterative algorithms such as Kleinman algorithm to solve the positive definite matrix P. According to the formula Calculate the optimal feedback gain matrix K at the current moment. Combine the gain matrix K with the global optimal state estimation vector Multiply, that is The obtained result is appropriately scaled and sent as a command to the servo motor of the force feedback steering wheel to drive it to generate the target output torque.

[0015] The operating conditions of the vehicle are constantly changing, which also interferes with the accuracy of the Kalman filter. To improve the accuracy of the Kalman filter, in a more specific embodiment, the process of determining the cubature Kalman filter and the measurement noise covariance matrix using the tire state and the steering wheel angular velocity includes: Calculating a tire saturation factor and an absolute value of a steering wheel angular velocity by combining the virtual sensor data, vehicle model parameters, and the global optimal state estimation vector; Taking the tire saturation factor and the absolute value of the steering wheel angular velocity as input, respectively calculating scaling coefficients of the process noise covariance matrix and the measurement noise covariance matrix through two nonlinear mapping functions; The scaling factor is multiplied by a preset reference process noise covariance matrix and a reference measurement noise covariance matrix respectively to obtain an adjusted process and measurement noise covariance matrix.

[0016] The process noise covariance matrix reflects the uncertainty of the vehicle model's predictions, while the measurement noise covariance matrix reflects the uncertainty of the sensor measurements. During aggressive driving, such as when tires approach their grip limits or the driver rapidly turns the steering wheel, the vehicle's nonlinear characteristics increase, significantly increasing the prediction error based on the linear model. Increasing the process noise covariance matrix reduces the filter's confidence in the model's predictions.

[0017] In one embodiment, the vehicle kinematic model calculates the slip angles of the front and rear tires using the center of mass slip angle and yaw rate provided in the global optimal state estimate vector, combined with the vehicle speed and real-time steering wheel angle from the virtual sensor data. The lateral forces currently acting on the front and rear tires are estimated based on these slip angles and the tire models pre-set in the vehicle model parameters. The estimated lateral forces are compared with the maximum lateral adhesion that the tires can provide under the current vertical load, and the ratio is the tire saturation factor. The maximum lateral adhesion is calculated, for example, from the road adhesion coefficient and the vertical load. Preferably, the absolute value of the steering wheel angular velocity is obtained by performing a temporal difference operation on the real-time steering wheel angle measurement and taking the absolute value.

[0018] For example, if the calculated tire saturation factor increases from 0.2 to 0.8, while the absolute value of the steering wheel angular velocity increases from 10 degrees per second to 200 degrees per second, the vehicle enters a nonlinear region. A pre-defined nonlinear mapping function, such as a sigmoid function, maps these two inputs to a larger process noise scaling factor, such as 5.0, while potentially adjusting the measurement noise scaling factor to 1.5. These two coefficients are multiplied by a pre-defined baseline noise covariance matrix, such as a diagonal matrix, to produce a dynamically adjusted covariance matrix. This allows the filter to maintain accurate state estimation performance under varying operating conditions.

[0019] In a more specific embodiment, the calculating of the confidence weight of each sub-model according to the vehicle speed and lateral acceleration in the virtual sensor data includes: Set the center point coordinates of vehicle speed and lateral acceleration for low speed, medium speed and high speed respectively; Calculate the Euclidean distance between the vehicle speed and lateral acceleration obtained by the virtual sensor data at the current moment and the center point coordinates of each working condition; Substituting the Euclidean distance into the Gaussian membership function, the confidence scores corresponding to the three sub-models are calculated; The three confidence levels are normalized so that the sum of the three confidence levels is 1, thereby obtaining a normalized confidence weight.

[0020] A single linear vehicle model cannot accurately describe all driving scenarios, from low speed to high speed and from linear to nonlinear. Therefore, this embodiment employs a multi-model switching strategy. By establishing independent sub-models for different typical operating conditions, the vehicle characteristics under each condition can be more accurately captured. To achieve a smooth transition between models, the applicability of each sub-model, i.e., the confidence weight, is calculated based on the current vehicle state. For example, assume that the coordinates of the center points for the three operating conditions are: 20 kilometers per hour and a lateral acceleration of 0.5 seconds per square meter for the low-speed zone; 70 kilometers per hour and a lateral acceleration of 2 seconds per square meter for the medium-speed-linear zone; and 120 kilometers per hour and a lateral acceleration of 6 seconds per square meter for the high-speed-nonlinear zone. When the virtual sensor measures a current vehicle speed of 75 kilometers per hour and a lateral acceleration of 2.5 seconds per square meter, the Euclidean distance from that point to the three center points is calculated. This point is closest to the center point of the medium-speed-linear zone and farthest from the low-speed zone. Using the Gaussian membership function, closer distances indicate higher membership. The initial confidence scores might be 0.1 for low speed, 0.8 for medium speed, and 0.3 for high speed. They are then normalized to 0.08 for low speed, 0.67 for medium speed, and 0.25 for high speed, with the sum of the three being 1.

[0021] In a more specific embodiment, the state estimation vectors output in parallel by each filter and the system matrix of each sub-model are weightedly fused using the confidence weights to obtain the global optimal state estimation vector and the equivalent linear system model for controller design, including: Extract the state matrices A1, A2, A3 and control matrices B1, B2, B3 from the low-speed, medium-speed, and high-speed state space submodels respectively; Using the calculated confidence weights w1, w2, and w3 corresponding to each sub-model, the state matrix of the equivalent linear system model is calculated using the following linear weighted summation formula: and control matrix : ; .

[0022] After obtaining the confidence weights of each sub-model under the current working conditions, in order to provide an accurate system model for the subsequent model predictive controller, the three discrete sub-models are fused into an equivalent global model. The fusion process adopts fuzzy logic and integrates the vehicle dynamic characteristics under different working conditions through weighted averaging, avoiding possible control mutations when switching models. Taking the weights calculated above as an example, w1 is equal to 0.08, w2 is equal to 0.67, and w3 is equal to 0.25. The state matrices A1, A2, A3 and control matrices B1, B2, B3 of the three pre-established sub-model libraries are retrieved respectively. Equivalent state matrix This is A1 multiplied by 0.08, plus A2 multiplied by 0.67, plus A3 multiplied by 0.25. Similarly, the equivalent control matrix The weighted summation is also performed in the same way. and The matrix integrates the characteristics of each sub-model under the current working conditions to obtain an equivalent linear time-varying model that can accurately reflect the current real-time response of the vehicle.

[0023] In a more specific embodiment, performing a short-time Fourier transform on the real-time steering wheel angle measurement value and using the high-frequency energy ratio as the driver input characteristic parameter includes: Set the time window and overlap rate to perform frame processing on the acquired steering wheel angle signal sequence; Apply a window function to each frame signal and perform a fast Fourier transform to obtain the frequency spectrum of the frame; Setting the frequency range of the high-frequency band and the full-frequency band, and calculating the sum of the spectrum energy in the high-frequency band and the sum of the energy in the full-frequency band; The ratio of the high-frequency energy sum to the full-band energy sum is used as the driver input characteristic parameter.

[0024] Different drivers have different driving styles, such as smooth versus aggressive. Drivers' steering habits are reflected in the steering wheel angle signal. Smooth steering maneuvers primarily contain low-frequency components, while rapid, abrupt steering movements generate a large number of high-frequency components. By analyzing the spectral characteristics of the steering wheel angle signal, the driver's instantaneous steering intensity can be identified. In practice, steering wheel angle data is continuously collected. For example, a 1-second window is set, sliding forward with a 50% overlap, meaning calculations are performed every 0.5 seconds. The data within each window, such as 256 samples, is first processed using a Hanning window to reduce spectral leakage, followed by a fast Fourier transform. Assume that frequencies above 2 Hz are defined as the high-frequency band, and 0 to 25 Hz as the full-frequency band. The sum of the energy in the 2 to 25 Hz range of the spectrum is calculated, followed by the total energy in the 0 to 25 Hz range. If the driver is making an emergency evasive maneuver, the calculated high-frequency energy ratio may be as high as 0.8; during steady cruising, this ratio may be less than 0.2.

[0025] In a more specific embodiment, determining the state weighting matrix and the control weighting matrix in the cost function of the linear quadratic regulator according to the three-dimensional lookup table includes: Construct a three-dimensional lookup table whose three dimensional axes correspond to vehicle speed, center of mass sideslip angle, and driver input characteristic parameters, and store pre-calibrated element values ​​of the state weight matrix and control weight matrix at discrete grid points within the three-dimensional lookup table; The vehicle speed component, the center of mass sideslip angle component and the calculated driver input characteristic parameters in the global optimal state estimation vector at the current moment are used as input coordinates; The trilinear interpolation algorithm is used to calculate the state weighting matrix and control weighting matrix under the current working condition according to the values ​​of the input coordinates at eight adjacent grid points in the lookup table.

[0026] The performance of a model predictive controller depends on the weighting matrix in its cost function. These matrices determine the balance between multiple objectives, such as path tracking accuracy, driving stability, and minimizing control energy consumption. Using a fixed weighting matrix cannot adapt to changing driving conditions and driver intent. For example, during high-speed emergency obstacle avoidance, greater emphasis should be placed on vehicle stability, with a higher penalty weight applied to the slip angle. During low-speed parking, greater emphasis should be placed on control smoothness, with a higher penalty weight applied to the control variable. To enable real-time adjustments, a three-dimensional lookup table is pre-built. For example, the speed axis ranges from 0 to 150 kilometers per hour, the slip angle axis ranges from -4 to +4 degrees, and the driver input characteristic parameter axis ranges from 0 to 1. At the intersection of these axes, weighting matrix values, optimized through extensive simulation and real-vehicle testing, are stored. When the vehicle is operating at a real-time speed of 95 kilometers per hour, a slip angle of 1.5 degrees, and a driver input characteristic parameter of 0.75, the vehicle locates the coordinate point in the lookup table and finds the eight nearest grid points. Through the trilinear interpolation algorithm, based on the relative distance between the current coordinate point and these eight grid points, a state and control weighting matrix tailored to the current specific working condition is calculated, thereby achieving online optimization of the controller performance.

[0027] In a second embodiment of the present invention, a dynamic force feedback closed-loop control system for simulated driving is provided, comprising the following modules: A configuration module is configured to obtain vehicle model parameters, real-time steering wheel angle measurements, and virtual sensor data output by a vehicle solver, including vehicle speed and lateral acceleration; based on the vehicle model parameters, establish state-space submodels corresponding to low-speed, medium-speed, and high-speed driving conditions, configure a corresponding cubic Kalman filter for each submodel, and determine the process and measurement noise covariance matrix of the cubic Kalman filter using tire state and steering wheel angular velocity; a model building module for calculating confidence weights for each sub-model based on the vehicle speed and lateral acceleration in the virtual sensor data, and using the confidence weights to perform weighted fusion on the state estimation vectors output in parallel by each filter and the system matrix of each sub-model to obtain a global optimal state estimation vector and an equivalent linear system model for controller design; a transformation module for performing a short-time Fourier transform on the real-time steering wheel angle measurement value, using the high-frequency energy ratio as a driver input characteristic parameter; constructing a linear quadratic regulator, and determining a state weighting matrix and a control weighting matrix in a cost function of the linear quadratic regulator based on a three-dimensional lookup table, wherein the inputs of the lookup table are the vehicle speed and center of mass sideslip angle components in the global optimal state estimation vector, and the driver input characteristic parameter; A control module is used to solve the algebraic Riccati equation based on the equivalent linear system model and the state weighting matrix and the control weighting matrix to obtain an optimal feedback gain matrix, and multiply the gain matrix with the global optimal state estimation vector to obtain a target output torque of the force feedback steering wheel actuator motor.

[0028] It should be noted that for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the embodiments of this specification are not limited by the order of the actions described, because according to the embodiments of this specification, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the embodiments of this specification.

[0029] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0030] The preferred embodiments disclosed above are intended only to help illustrate this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and variations can be made based on the content of the embodiments of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of this specification, so that those skilled in the art can better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. A dynamic force feedback closed-loop control method in simulated driving, characterized in that: The following steps are involved: Obtain vehicle model parameters, real-time steering wheel angle measurements, and virtual sensor data including vehicle speed and lateral acceleration output by the vehicle solver; Based on the vehicle model parameters, state space sub-models corresponding to low-speed, medium-speed, and high-speed driving conditions are established, and a corresponding cubic Kalman filter is configured for each sub-model. The process and measurement noise covariance matrix of the cubic Kalman filter is determined using tire state and steering wheel angular velocity; Calculating confidence weights for each sub-model based on the vehicle speed and lateral acceleration in the virtual sensor data, and using the confidence weights to perform weighted fusion on the state estimation vectors output in parallel by each filter and the system matrix of each sub-model to obtain a global optimal state estimation vector and an equivalent linear system model; Perform short-time Fourier transform on the real-time steering wheel angle measurement and use the high-frequency energy ratio as the driver input feature parameter; Constructing a linear quadratic regulator and determining a state weighting matrix and a control weighting matrix in a cost function of the linear quadratic regulator based on a three-dimensional lookup table, wherein the inputs of the lookup table are the vehicle speed and center of mass sideslip angle components in the global optimal state estimation vector and the driver input characteristic parameter; The algebraic Riccati equation is solved based on the equivalent linear system model, the state weighting matrix and the control weighting matrix to obtain the optimal feedback gain matrix, and the gain matrix is ​​multiplied by the global optimal state estimation vector to obtain the target output torque of the force feedback steering wheel actuator motor.

2. The method according to claim 1, characterized in that The process of determining the cubature Kalman filter and the measurement noise covariance matrix using the tire state and the steering wheel angular velocity includes: Calculating a tire saturation factor and an absolute value of a steering wheel angular velocity by combining the virtual sensor data, vehicle model parameters, and the global optimal state estimation vector; Taking the tire saturation factor and the absolute value of the steering wheel angular velocity as input, respectively calculating scaling coefficients of the process noise covariance matrix and the measurement noise covariance matrix through two nonlinear mapping functions; The scaling factor is multiplied by a preset reference process noise covariance matrix and a reference measurement noise covariance matrix respectively to obtain an adjusted process and measurement noise covariance matrix.

3. The method according to claim 1, characterized in that The calculating of the confidence weight of each sub-model according to the vehicle speed and lateral acceleration in the virtual sensor data includes: Set the center point coordinates of vehicle speed and lateral acceleration for low speed, medium speed and high speed respectively; Calculate the Euclidean distance between the vehicle speed and lateral acceleration obtained by the virtual sensor data at the current moment and the center point coordinates of each working condition; Substituting the Euclidean distance into the Gaussian membership function, the confidence scores corresponding to the three sub-models are calculated; The three confidence levels are normalized so that the sum of the three confidence levels is 1, thereby obtaining a normalized confidence weight.

4. The method according to claim 1, wherein The method uses the confidence weights to perform weighted fusion on the state estimation vectors output in parallel by each filter and the system matrix of each sub-model to obtain a global optimal state estimation vector and an equivalent linear system model for controller design, including: Extract the state matrices A1, A2, A3 and control matrices B1, B2, B3 from the low-speed, medium-speed, and high-speed state space submodels respectively; Using the calculated confidence weights w1, w2, and w3 corresponding to each sub-model, the state matrix of the equivalent linear system model is calculated using the following linear weighted summation formula: and control matrix : ； 。 5. The method according to claim 1, wherein The method of performing a short-time Fourier transform on the real-time steering wheel angle measurement value and using the high-frequency energy ratio as the driver input characteristic parameter includes: Set the time window and overlap rate to perform frame processing on the acquired steering wheel angle signal sequence; Apply a window function to each frame signal and perform a fast Fourier transform to obtain the frequency spectrum of the frame; Setting the frequency range of the high-frequency band and the full-frequency band, and calculating the sum of the spectrum energy in the high-frequency band and the sum of the energy in the full-frequency band; The ratio of the high-frequency energy sum to the full-band energy sum is used as the driver input characteristic parameter.

6. The method according to claim 1, characterized in that The determining of the state weighting matrix and the control weighting matrix in the cost function of the linear quadratic regulator according to the three-dimensional lookup table includes: Construct a three-dimensional lookup table whose three dimensional axes correspond to vehicle speed, center of mass sideslip angle, and driver input characteristic parameters, and store pre-calibrated element values ​​of the state weight matrix and control weight matrix at discrete grid points within the three-dimensional lookup table; The vehicle speed component, the center of mass sideslip angle component and the calculated driver input characteristic parameters in the global optimal state estimation vector at the current moment are used as input coordinates; The trilinear interpolation algorithm is used to calculate the state weighting matrix and control weighting matrix under the current working condition according to the values ​​of the input coordinates at eight adjacent grid points in the lookup table.

7. A dynamic force feedback closed-loop control system for simulated driving, characterized in that: Includes the following modules: A configuration module is configured to obtain vehicle model parameters, real-time steering wheel angle measurements, and virtual sensor data output by a vehicle solver, including vehicle speed and lateral acceleration; based on the vehicle model parameters, establish state-space submodels corresponding to low-speed, medium-speed, and high-speed driving conditions, configure a corresponding cubic Kalman filter for each submodel, and determine the process and measurement noise covariance matrix of the cubic Kalman filter using tire state and steering wheel angular velocity; a model building module for calculating confidence weights for each sub-model based on the vehicle speed and lateral acceleration in the virtual sensor data, and using the confidence weights to perform weighted fusion on the state estimation vectors output in parallel by each filter and the system matrix of each sub-model to obtain a global optimal state estimation vector and an equivalent linear system model; A transformation module is used to perform short-time Fourier transform on the real-time steering wheel angle measurement value and use the high-frequency energy ratio as the driver input feature parameter; Constructing a linear quadratic regulator and determining a state weighting matrix and a control weighting matrix in a cost function of the linear quadratic regulator based on a three-dimensional lookup table, wherein the inputs of the lookup table are the vehicle speed and center of mass sideslip angle components in the global optimal state estimation vector and the driver input characteristic parameter; A control module is used to solve the algebraic Riccati equation based on the equivalent linear system model and the state weighting matrix and the control weighting matrix to obtain an optimal feedback gain matrix, and multiply the gain matrix with the global optimal state estimation vector to obtain a target output torque of the force feedback steering wheel actuator motor.

8. The system according to claim 7, characterized in that The process of determining the cubature Kalman filter and the measurement noise covariance matrix using the tire state and the steering wheel angular velocity includes: Calculating a tire saturation factor and an absolute value of a steering wheel angular velocity by combining the virtual sensor data, vehicle model parameters, and the global optimal state estimation vector; Taking the tire saturation factor and the absolute value of the steering wheel angular velocity as input, respectively calculating scaling coefficients of the process noise covariance matrix and the measurement noise covariance matrix through two nonlinear mapping functions; The scaling factor is multiplied by a preset reference process noise covariance matrix and a reference measurement noise covariance matrix respectively to obtain an adjusted process and measurement noise covariance matrix.

9. The system according to claim 7, wherein: The calculating of the confidence weight of each sub-model according to the vehicle speed and lateral acceleration in the virtual sensor data includes: Set the center point coordinates of vehicle speed and lateral acceleration for low speed, medium speed and high speed respectively; Calculate the Euclidean distance between the vehicle speed and lateral acceleration obtained by the virtual sensor data at the current moment and the center point coordinates of each working condition; Substituting the Euclidean distance into the Gaussian membership function, the confidence scores corresponding to the three sub-models are calculated; The three confidence levels are normalized so that the sum of the three confidence levels is 1, thereby obtaining a normalized confidence weight.

10. The system according to claim 7, wherein: The method uses the confidence weights to perform weighted fusion on the state estimation vectors output in parallel by each filter and the system matrix of each sub-model to obtain a global optimal state estimation vector and an equivalent linear system model for controller design, including: Extract the state matrices A1, A2, A3 and control matrices B1, B2, B3 from the low-speed, medium-speed, and high-speed state space submodels respectively; Using the calculated confidence weights w1, w2, and w3 corresponding to each sub-model, the state matrix of the equivalent linear system model is calculated using the following linear weighted summation formula: and control matrix : ； 。

Images