Steering wheel torque prediction method and device, server and storage medium

Abstract

The embodiment of the invention provides a steering wheel torque prediction method and device, a server and a storage medium, and relates to the technical field of steering-by-wire system development and testing. Real vehicle data can be obtained, and the real vehicle data comprises the steering wheel angle, the angular velocity, the vehicle speed, the lateral acceleration, the vertical load and the steering wheel torque of the vehicle; determining a steering wheel torque prediction model according to the real vehicle data; and predicting the steering wheel torque by using the steering wheel torque prediction model. The method comprehensively considers the influence of various factors such as the dynamic characteristics of the steering system and the road condition on the road feeling feedback effect, can improve the accuracy of the steering wheel torque prediction model for predicting the steering wheel torque, and improves the test efficiency and evaluation accuracy of the steer-by-wire system.

Description

Technical Field

[0001] The present application relates to the technical field of development and testing of wire-controlled steering systems, and more specifically, to a steering wheel torque prediction method, device, server and storage medium. Background Art

[0002] With the development of intelligent connected vehicle technology, the steer-by-wire (SBW) system has become an important development direction for future automotive steering systems due to its flexibility and customizability. With the continuous development and maturity of SBW technology, more and more vehicle manufacturers and parts suppliers have begun to develop and produce SBW systems. In order to ensure that the performance of the SBW system meets the design requirements, a lot of testing and verification work needs to be done during the development process. Among them, the road feel feedback effect is one of the important indicators for evaluating the performance of the SBW system. At present, the industry generally adopts a combination of bench tests and real vehicle tests to test the road feel feedback effect of the SBW system. However, this traditional testing method has the disadvantages of long cycle, high cost, and limited coverage of working conditions, making it difficult to fully evaluate the performance of the SBW system under various complex working conditions. Summary of the invention

[0003] The embodiments of the present application propose a steering wheel torque prediction method, device, server and storage medium to solve the above technical problems.

[0004] In a first aspect, an embodiment of the present application provides a steering wheel torque prediction method, the method comprising: acquiring real vehicle data, the real vehicle data comprising the vehicle's steering wheel angle, angular velocity, vehicle speed, lateral acceleration, vertical load and steering wheel torque; determining a steering wheel torque prediction model based on the real vehicle data; and predicting the steering wheel torque using the steering wheel torque prediction model.

[0005] In the second aspect, an embodiment of the present application provides a steering wheel torque prediction device, which includes: a data acquisition module for acquiring real vehicle data, the real vehicle data including the vehicle's steering wheel angle, angular velocity, vehicle speed, lateral acceleration, vertical load and steering wheel torque; a model training module for determining a steering wheel torque prediction model based on the real vehicle data; and a torque prediction module for predicting the steering wheel torque using the steering wheel torque prediction model.

[0006] In a third aspect, an embodiment of the present application provides a server, comprising: a memory and a processor, wherein an application is stored in the memory, and the application is used to enable the processor to execute the method provided by the embodiment of the present application when called by the processor.

[0007] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a program code stored thereon, wherein the program code is used to enable the processor to execute the method provided by the embodiment of the present application when called by the processor.

[0008] The steering wheel torque prediction method provided in the embodiment of the present application has the following technical effects: according to the steering wheel angle, angular velocity, vehicle speed, lateral acceleration, vertical load and steering wheel torque collected from the actual vehicle, a steering wheel torque prediction model is determined, and the influence of various factors such as the dynamic characteristics of the steering system and the road conditions on the road feel feedback effect is comprehensively considered, so that the accuracy of the steering wheel torque prediction model in predicting the steering wheel torque can be improved. In addition, the steering wheel torque is predicted using the steering wheel torque prediction model, and the predicted steering wheel torque can simulate the road feel feedback effect of the wire steering system, so that the road feel feedback effect of the steering wheel torque prediction model (i.e., the wire steering system) can be evaluated and verified based on the predicted steering wheel torque in the future, reducing the number of experiments and improving the test efficiency and evaluation accuracy of the wire steering system. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following is a brief introduction to the drawings required for the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments and drawings obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0010] Figure 1 A schematic diagram showing a hardware environment for developing and testing a steer-by-wire system provided in an embodiment of the present application;

[0011] Figure 2 A schematic diagram of a vehicle-to-cloud communication architecture provided in an embodiment of the present application is shown;

[0012] Figure 3 A schematic diagram showing a process flow of developing and testing a steer-by-wire system provided in an embodiment of the present application is shown;

[0013] Figure 4 A schematic diagram of a process flow of a steering wheel torque prediction method provided by an embodiment of the present application is shown;

[0014] Figure 5 A schematic diagram showing a flow chart of a steering wheel torque prediction method provided by another embodiment of the present application is shown;

[0015] Figure 6 A schematic diagram of the process of step S240 provided in another embodiment of the present application is shown;

[0016] Figure 7A schematic diagram showing a flow chart of a steering wheel torque prediction method provided by another embodiment of the present application is shown;

[0017] Figure 8 A schematic diagram of the structure of a steering wheel torque prediction device provided in an embodiment of the present application is shown;

[0018] Fig. 9 A schematic diagram of the structure of a server provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0019] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0020] Here, the terms involved in the embodiments of the present application are explained:

[0021] Road feel feedback: Road feel feedback refers to the information from the road and vehicle dynamics that the driver feels through the steering wheel. This feedback is critical to the driver's sense of control and the stability of the vehicle. In a mechanical steering system, this feedback is transmitted to the steering wheel through direct contact between the wheel and the road. In a wire-controlled steering system, since there is no mechanical connection between the steering wheel and the wheel, road feel needs to be simulated and provided by electronic sensors and control algorithms.

[0022] Feedback torque: Feedback torque refers to the torque used to simulate road feel and transmitted to the driver in a wire-controlled steering system. This torque is generated by the motor according to the control algorithm, and the purpose is to make the driver feel force feedback similar to that of a traditional steering system. The control algorithm calculates the appropriate feedback torque based on input parameters such as vehicle speed, steering angle, and the motion state of the suspension system.

[0023] This application uses two evaluation indicators for each working condition: mean absolute error and mean square error.

[0024] Mean Absolute Error (MAE): MAE refers to the average of the absolute value of the difference between the predicted value (i.e., the steering wheel torque predicted by the steering wheel torque prediction model) and the true value (i.e., the steering wheel torque actually collected).

[0025] Mean Squared Error (MSE): MSE refers to the average of the squares of the differences between the predicted values ​​and the true values.

[0026] N·m stands for Newton·meter, which is the unit of torque. In the embodiment of the present application, it represents the unit of steering wheel torque.

[0027] (N·m) 2Represents Newton meter squared, which is the unit of MSE.

[0028] Out-of-Distribution Generalization (OOD): OOD refers to the ability of a machine learning model to process data with a distribution that is different from its training data. It is verified by evaluating the performance of the model on unseen data distributions.

[0029] OOD error: OOD error refers to the average prediction error of the model on each OOD test set, reflecting the overall level of the model's generalization ability.

[0030] OOD robustness: OOD robustness refers to the increment of the model’s prediction error on the OOD test set relative to the original test set, reflecting the model’s sensitivity to distribution shift.

[0031] Performance under extreme conditions: Performance under extreme conditions refers to the prediction error and failure rate of the model under extreme conditions, reflecting the reliability of the model under the worst conditions.

[0032] See also Figure 1 , Figure 1 A schematic diagram of the hardware environment for the development and testing of the wire control steering system provided in an embodiment of the present application is shown. The hardware environment 100 includes a vehicle 110 and a server 120. The vehicle 110 may include but is not limited to a gasoline vehicle or an electric vehicle, and the electric vehicle may include but is not limited to a pure electric vehicle, a hybrid vehicle, or a fuel cell vehicle. The server 120 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0033] See also Figure 2 , Figure 2 FIG. 1 is a schematic diagram of a vehicle-to-cloud communication architecture provided by an embodiment of the present application. Figure 2 As shown, vehicle 110 includes a telematics box (TBOX) 111 and a wire-controlled steering system 112. In a real vehicle or bench test, the steering wheel torque prediction model can be run on a remote or local server. For example, server 120 is a cloud server (referred to as the cloud). Figure 2As shown in the figure, the on-board TBOX acts as a network proxy and communicates with the cloud through the 4G / 5G network. When testing the SBW system, the test command is sent from the cloud to the TBOX through the Test Service Protocol (TSP), and then the TBOX passes it to the steer-by-wire system through the Controller Area Network (CAN) bus. The steer-by-wire system executes the command and uploads the sensor data to the TBOX, which then transmits it back to the cloud for use by the steering wheel torque prediction model.

[0034] See also Figure 3 , Figure 3 FIG. 1 is a flow chart showing the development and testing process of the wire control steering system provided by the embodiment of the present application. Figure 3 As shown, when developing and testing the wire-controlled steering system, real vehicle data can be collected and preprocessed first (S1), and then the data set can be managed (S2). Generate a generalized test set (i.e., OOD test set) based on the real vehicle data (S3). Determine the steering wheel torque prediction model based on the real vehicle data (S4), including: model structure construction (S41), model training (S42), and generalization ability optimization (S43). After that, perform model evaluation and iterative optimization (S5), including: model evaluation (S51), iterative optimization (S52), and model version management (S53). Finally, perform experimental verification (S6), including: bench test (S61), real vehicle test (S62), and performance comparison (S63). Figure 3 For a detailed description of each process shown, please refer to the relevant sections below.

[0035] The steering wheel torque prediction method in the embodiment of the present application can be applied to a server or a steering wheel torque prediction device, and the steering wheel torque prediction device can be integrated in the server. Next, the steering wheel torque prediction method will be described in detail.

[0036] See also Figure 4 , Figure 4 The flowchart of the steering wheel torque prediction method provided by an embodiment of the present application is shown. The steering wheel torque prediction method may include steps S110 to S130.

[0037] Step S110: Acquire real vehicle data, where the real vehicle data includes the vehicle's steering wheel angle, angular velocity, vehicle speed, lateral acceleration, vertical load, and steering wheel torque.

[0038] Real vehicle data refers to the actual data collected when the vehicle is actually running under various working conditions. Real vehicle data may include but is not limited to the vehicle's steering wheel angle θ, angular velocity ω, vehicle speed v, lateral acceleration a, vertical load F, and corresponding steering wheel torque T. Each item of data in the real vehicle data (such as the steering wheel angle) can be called a feature. Data acquisition equipment can be deployed on the real vehicle, and a sufficient amount of real vehicle data can be collected under various working conditions (for example, straight lines, curves, high and low speeds, acceleration and deceleration, and slope crossing) through the data acquisition equipment. The cloud can collect real vehicle data through vehicle-cloud communication.

[0039] In some embodiments, after the real vehicle data is collected, the collected original real vehicle data can be preprocessed. Specifically, the collected original real vehicle data can be cleaned and feature extracted. For example, the outliers in the original real vehicle data are eliminated, the noise is smoothed, and the appropriate feature combination (for example, the absolute value of the steering wheel angle, the sign of the angular velocity, etc.) is selected according to the physical meaning and prior knowledge. The preprocessed real vehicle data is randomly divided into a training set, a validation set, and a test set according to a certain ratio. For example, according to the ratio of 8:1:1, the training set, the validation set, and the test set are randomly divided. Assume that after preprocessing, 10,000 data are obtained, including 8,000 training sets, 1,000 validation sets, and 1,000 test sets.

[0040] In some embodiments, when dividing the training set, validation set and test set, they can be divided according to a specific data set division protocol, so as to facilitate the management and testing of the performance of the wire control steering system. For example, the pre-processed real vehicle data can be managed according to directories, and each directory corresponds to a working condition or parameter. For example, the vehicle speed directory is divided into sub-directories such as low speed, medium speed, and high speed. Each sub-directory contains multiple data files, recording signals such as vehicle speed, acceleration, load, steering wheel angle, angular velocity and torque over a period of time. Users can customize sub-directories as needed to flexibly manage data under different working conditions. Exemplarily, it can be divided in the manner shown in Table 1. The use of the vehicle-cloud-vehicle communication architecture and the data set division protocol realizes the efficient management and utilization of massive data, providing data support for the performance evaluation and verification of the SBW system.

[0041] Table 1

[0042] Division Division results By speed Low speed, medium speed, medium-high speed, high speed, generalized high speed By steering angle Small angle, medium angle, large angle By lateral acceleration Small acceleration, medium acceleration, large acceleration By load Empty, half loaded, fully loaded By road type Straight lines, curves, ramps, bumpy roads, etc.

[0043] Step S120: Determine a steering wheel torque prediction model based on the actual vehicle data.

[0044] This application is based on a nonlinear mathematical model containing quadratic terms and cross terms. In order to describe the complex relationship between the road feel torque and the influencing factors, a steering wheel torque prediction model (general form) is constructed as shown in the following expression:

[0045] T=k0+k1·θ+k2·ω+k3·v+k4·a+k5·F+k6·θ 2 +k7·ω 2 +k8·v 2 +k9·a 2 +k 10 ·F 2 +k 11 ·θ·ω+k 12 ·θ·v+k 13 ·θ·a+k 14 ·θ·F+k 15 ·ω·v+k 16 ·ω·a+k 17 ·ω·F+k 18 ·v·a+k 19 ·v·F+k 20 ·a·F

[0046] Where T is the steering wheel torque, θ is the steering wheel angle, ω is the angular velocity, v is the vehicle speed, a is the lateral acceleration, F is the vertical load, k0 to k 20 are the model parameters to be fitted; according to the actual vehicle data (training set), the model parameters in the steering wheel torque prediction model are fitted to obtain the steering wheel torque prediction model.

[0047] In some embodiments, the model parameters k0 to k1 in the steering wheel torque prediction model can be fitted using the least squares method according to the actual vehicle data (training set). 20 , and obtain the steering wheel torque prediction model.

[0048] As an example, assume that there are n samples in the training set, each of which contains 5 features (θ, ω, v, a, F) and 1 target value (T). The training set can be represented in matrix form:

[0049]

[0050] Where X is an n×21 feature matrix, each row corresponds to a sample, the first column is all 1 (used to fit the bias term k0), and the following 20 columns are the linear terms, quadratic terms, and cross terms of θ, ω, v, a, and F respectively; Y is an n×1 target value vector.

[0051] The goal is to find a set of parameters Make X k The error with Y is the smallest. According to the least squares method, the optimal parameter k satisfies: k = (X T X) -1 X T Y, where X T is the transposed matrix of X, (X TX) -1 Yes X T The inverse matrix of X. Substituting the feature matrix X of the training set and the target value vector Y into the above formula, the estimated value of the model parameter k can be obtained. Assume that the fitted parameters are as follows: k0 = 0.5, k1 = 2.1, k2 = 0.8, k3 = 0.02, k4 = 1.5, k5 = 0.3, k6 = 0.1, k7 = 0.05, k8 = 0.001, k9 = 0.2, k 10 =0.01, k 11 =-0.3, k 12 =0.02, k 13 =-0.1, k 14 =0.01, k 15 =-0.02, k 16 =0.1, k 17 =-0.01, k 18 =0.05, k 19 =0.005, k 20 =0.03.

[0052] In some other embodiments, the model parameters k0 to k in the steering wheel torque prediction model can be fitted using the gradient descent method according to the actual vehicle data (training set). 20 , and obtain the steering wheel torque prediction model.

[0053] The model parameters k0 to k in the fitted steering wheel torque prediction model 20 In the process of model training, the mean absolute error (MAE), mean square error (MSE), R 2 At least one of the above is used as a loss function, the loss of each fitting is calculated, and based on the loss, it is determined whether to continue fitting or stop fitting. For example, it can be determined whether the loss is less than the preset loss; if the loss is less than the preset loss, it is determined that the fitting effect is good and the accuracy of the steering wheel torque prediction model is high, and the fitting can be stopped to obtain the steering wheel torque prediction model; if the loss is not less than the preset loss, it is determined that the fitting effect is poor and the accuracy of the steering wheel torque prediction model is low, and the fitting can continue until the loss is less than the preset loss.

[0054] In some embodiments, the validation set data can be used to evaluate the fitting effect of the steering wheel torque prediction model and calculate the MSE, MAE, R 2 At the same time, draw the true torque curve and the predicted torque curve to visually check the fitting quality. For example, suppose that on the validation set, the mean absolute error (MAE) of the steering wheel torque prediction model is 0.3N·m and the mean square error (MSE) is 0.2(N·m) 2 This shows that the model can fit the validation set data well.

[0055] Step S130: using the steering wheel torque prediction model to predict the steering wheel torque.

[0056] The steering wheel torque prediction model has good model performance. The steering wheel torque prediction model can be deployed as a new version to a local or cloud control unit to replace the original verification model, so as to use the steering wheel torque prediction model to predict the steering wheel torque.

[0057] Steps S110 to S130 have the following technical effects: according to the steering wheel angle, angular velocity, vehicle speed, lateral acceleration, vertical load and steering wheel torque collected from the actual vehicle, a steering wheel torque prediction model is determined, and the influence of various factors such as the dynamic characteristics of the steering system and the road conditions on the road feel feedback effect is comprehensively considered, so that the accuracy of the steering wheel torque prediction model in predicting the steering wheel torque can be improved. In addition, the steering wheel torque is predicted using the steering wheel torque prediction model, and the predicted steering wheel torque can simulate the road feel feedback effect of the wire steering system, so that the road feel feedback effect of the steering wheel torque prediction model (i.e., the wire steering system) can be evaluated and verified based on the predicted steering wheel torque in the future, reducing the number of experiments and improving the test efficiency and evaluation accuracy of the wire steering system.

[0058] See also Figure 5 , Figure 5 A schematic flow chart of a steering wheel torque prediction method provided by another embodiment of the present application is shown. The steering wheel torque prediction method may include steps S210 to S250.

[0059] Step S210: Acquire real vehicle data, where the real vehicle data includes the vehicle's steering wheel angle, angular velocity, vehicle speed, lateral acceleration, vertical load, and steering wheel torque.

[0060] For a detailed description of step S210 , please refer to step S110 .

[0061] Step S220: Determine a steering wheel torque prediction model based on the actual vehicle data.

[0062] For a detailed description of step S220 , please refer to step S120 .

[0063] Step S230: adding multiple penalty items to the steering wheel torque prediction model to obtain an objective function, wherein the multiple penalty items include error penalty for out-of-distribution generalized data, extreme value penalty, and smoothness penalty.

[0064] In order to further improve the generalization ability of the steering wheel torque prediction model, the present application introduces multiple penalty terms in the steering wheel torque prediction model (general form), thereby explicitly constraining the performance of the steering wheel torque prediction model under OOD conditions. In an embodiment of the present application, the multiple penalty terms may include:

[0065] (1) Error penalty for out-of-distribution generalized data (OOD error penalty): The OOD error penalty refers to the prediction error of the steering wheel torque prediction model on out-of-distribution generalized data (OOD data). Introducing the prediction error term of OOD data allows the steering wheel torque prediction model to fit the original real vehicle data while also taking into account the performance of OOD data.

[0066] (2) Extreme value penalty: The extreme value penalty refers to the penalty term when the predicted value of the steering wheel torque prediction model under extreme working conditions exceeds the reasonable range. The introduction of extreme value penalty can limit the predicted value of the steering wheel torque prediction model under extreme working conditions to not exceed the reasonable range.

[0067] (3) Smoothness penalty: Smoothness penalty refers to the smoothness penalty term of the prediction curve of the steering wheel torque prediction model. The introduction of smoothness penalty can make the prediction curve of the steering wheel torque prediction model under OOD conditions smoother and avoid violent fluctuations.

[0068] By adding multiple penalty terms to the steering wheel torque prediction model, the objective function can be obtained as shown in the following expression:

[0069] T=k0+k1·θ+k2·ω+k3·v+k4·a+k5·F+k6·θ 2 +k7·ω 2 +k8·v 2 +k9·a 2 +k 10 ·F 2 +k 11 ·θ·ω+k 12 ·θ·v+k 13 ·θ·a+k 14 ·θ·F+k 15 ·ω·v+k 16 ·ω·a+k 17 ·ω·F+k 18 ·v·a+k 19 ·v·F+k 20 ·a·F+λ1·L OOD +λ2·R extreme +λ3·R smooth

[0070] Among them, L OOD represents the OOD error penalty, R extreme represents the extreme value penalty, R smooth represents smoothing penalty, λ1, λ2, and λ3 are balancing coefficients used to control the weights of multiple penalty items.

[0071] By introducing OOD error penalty, extreme value penalty and smoothness penalty in the objective function, the generalization ability of the steering wheel torque prediction model can be explicitly constrained during training, so that it can better adapt to OOD conditions and extreme working conditions while fitting the original real vehicle data. By introducing indicators such as OOD error, OOD robustness, and extreme working condition performance, the generalization ability of the model can be evaluated from multiple angles.

[0072] Among them, in order to determine the optimal value of the penalty weight coefficient, the cross-validation method can be used to select λ1, λ2, and λ3. The specific steps are as follows:

[0073] (1) Determination of the range of candidate values: The penalty weight coefficient is usually between 0 and 1. A smaller value means that the penalty has a weaker impact, while a larger value means that the penalty has a stronger impact. Based on experience and prior knowledge, a reasonable range of values ​​can be selected.

[0074] (2) Setting the candidate value interval: In order to compare the performance differences of the models under different weight coefficients, a certain interval should be maintained between the candidate values. If the interval is too small, it may be difficult to distinguish the model performance, and if the interval is too large, the optimal value may be missed. The candidate values ​​can be set in a logarithmic interval or geometric interval manner.

[0075] (3) Selection of the number of candidate values: The number of candidate values ​​should not be too large to avoid increasing the amount of calculation, nor too small to avoid being unable to find the optimal value. Generally, 3 to 5 candidate values ​​can be selected.

[0076] Assume that the given candidate value ranges are λ1∈{0.001, 0.01, 0.1}, λ2∈{0.01, 0.1, 1}, and λ3∈{0.005, 0.05, 0.5}. The optimal combination of penalty weight coefficients λ1, λ2, and λ3 can be selected by cross-validation. 3×3×3=27 different combinations need to be compared, and for each combination, the average performance index (such as MSE) of the model is calculated using 5-fold cross-validation.

[0077] Assume that after cross-validation, we get the following combination with the best performance: λ1=0.01, λ2=0.1, λ3=0.05. This means that when the penalty weight coefficient takes these values, the average performance of the model on the validation set is the best. The meanings of λ1=0.01, λ2=0.1, λ3=0.05 are as follows:

[0078] λ1=0.01 means that the weight of the prediction error of OOD data in the loss function is 0.01. This value is small, indicating that this application hopes that the model will consider the prediction effect of OOD data to a certain extent while fitting the training data, but does not want this item to dominate the entire loss function.

[0079] λ2=0.1 means that the weight of the extreme value penalty term in the loss function is 0.1. This value is one order of magnitude larger than λ1, indicating that this application pays more attention to the performance of the model under extreme working conditions, and hopes that the predicted value of the model will not have abnormal maximum or minimum values.

[0080] λ3=0.05 means that the weight of the smoothness penalty term in the loss function is 0.05. This value is between λ1 and λ2, indicating that this application hopes that the prediction curve of the model under OOD conditions should be as smooth as possible to avoid drastic fluctuations, but this requirement is not as important as the extreme value penalty term.

[0081] Step S240: Evaluate the generalization ability of the steering wheel torque prediction model according to the real vehicle data and the objective function.

[0082] See also Figure 6 , Figure 6 A schematic flow chart of step S240 provided in another embodiment of the present application is shown. Step S240 may include steps S241 to S243.

[0083] Step S241: performing distribution shift on the real vehicle data to obtain distribution out-of-distribution generalized data.

[0084] In order to comprehensively evaluate the generalization ability of the steering wheel torque prediction model, this application adopts a distribution shift method to generate a series of out-of-distribution generalization data (which can be called OOD test set) that obey different distributions based on the training set. Specifically, the distribution shift operation can be performed on the training set in at least one of the following ways to obtain the out-of-distribution generalization data:

[0085] (1) Mean shift: A random shift is applied to the real vehicle data to obtain mean shift data. For example, based on the real vehicle data used as a training set, a random shift can be applied to each feature to change the mean of the real vehicle data, thereby simulating vehicle parameter shift, sensor deviation, etc.

[0086] (2) Variance shift: A random scaling factor is applied to the real vehicle data to obtain variance shift data. For example, a random scaling factor can be applied to each feature based on the real vehicle data used as a training set to change the variance of the real vehicle data, thereby simulating working condition fluctuations, driving habit differences, and the like.

[0087] (3) Extreme value shift: The extreme value of the real vehicle data is enlarged or reduced to obtain extreme value shift data. For example, based on the real vehicle data used as the training set, the extreme values ​​of certain features can be further enlarged or reduced to simulate extreme working conditions such as sharp turns, emergency braking, and bumpy roads.

[0088] In some embodiments, the above three offset methods can be combined to generate more complex distribution-outside generalization data, thereby comprehensively testing the robustness of the steering wheel torque prediction model. For example, a random offset can be applied to the real vehicle data to obtain mean offset data; a random scaling factor can be applied to the real vehicle data to obtain variance offset data; the real vehicle data can be amplified or reduced by extreme values ​​to obtain extreme value offset data; the mean offset data, the variance offset data, and the extreme value offset data can be combined to obtain distribution-outside generalization data.

[0089] Step S242: inputting the out-of-distribution generalized data into the objective function to obtain the steering wheel torque predicted by the objective function.

[0090] By inputting the out-of-distribution generalization data into the objective function, the prediction performance of the steering wheel torque prediction model can be evaluated on the out-of-distribution generalization data (i.e., the OOD test set), thereby focusing on the generalization ability of the steering wheel torque prediction model under extreme working conditions and abnormal situations.

[0091] Step S243: evaluating the generalization ability of the steering wheel torque prediction model according to the error between the steering wheel torque predicted by the objective function and the steering wheel torque collected from the actual vehicle.

[0092] The error between the steering wheel torque predicted by the objective function and the steering wheel torque collected by the actual vehicle can be calculated, and the generalization ability of the steering wheel torque prediction model can be evaluated based on the error. The error can include MAE error, MSE error or R 2 The error can reflect the generalization ability of the steering wheel torque prediction model. If the error is small, it can be determined that the generalization ability of the steering wheel torque prediction model is good. If the error is large, it can be determined that the generalization ability of the steering wheel torque prediction model is poor.

[0093] Step S250: using the steering wheel torque prediction model to predict the steering wheel torque.

[0094] For a detailed description of step S250 , please refer to step S130 .

[0095] Compared with steps S110 to S130, steps S210 to S250 also have the following additional technical effects: by introducing OOD error penalty, extreme value penalty and smoothness penalty in the objective function, the generalization ability of the steering wheel torque prediction model can be explicitly constrained during training, so that it can better adapt to OOD conditions and extreme working conditions while fitting the original real vehicle data. The prediction performance of the steering wheel torque prediction model is evaluated on out-of-distribution generalized data (i.e., OOD test set), thereby focusing on the generalization ability of the steering wheel torque prediction model under extreme working conditions and abnormal conditions. In other words, the steering wheel torque prediction model optimized based on the OOD test set and penalty term in this application improves the prediction accuracy, generalization ability and real-time performance of the model compared to traditional linear models and simple nonlinear models.

[0096] See also Figure 7 , Figure 7 A schematic flow chart of a steering wheel torque prediction method provided by another embodiment of the present application is shown. The steering wheel torque prediction method may include steps S310 to S370.

[0097] Step S310: Acquire real vehicle data, where the real vehicle data includes the vehicle's steering wheel angle, angular velocity, vehicle speed, lateral acceleration, vertical load, and steering wheel torque.

[0098] For a detailed description of step S310 , please refer to step S110 .

[0099] Step S320: Determine a steering wheel torque prediction model based on the actual vehicle data.

[0100] For a detailed description of step S320 , please refer to step S120 .

[0101] Step S330: adding multiple penalty items to the steering wheel torque prediction model to obtain an objective function, wherein the multiple penalty items include error penalty for out-of-distribution generalized data, extreme value penalty, and smoothness penalty.

[0102] For a detailed description of step S330 , please refer to step S230 .

[0103] Step S340: Evaluate the generalization ability of the steering wheel torque prediction model according to the real vehicle data and the objective function.

[0104] For a detailed description of step S340 , please refer to step S240 .

[0105] Step S350: When the evaluation result does not meet the design requirements, iteratively optimize the steering wheel torque prediction model.

[0106] When the error between the steering wheel torque predicted by the objective function and the steering wheel torque collected from the actual vehicle is greater than the preset error, it can be determined that the evaluation result does not meet the design requirements. At this time, the shortcomings of the steering wheel torque prediction model can be analyzed to improve the model structure, feature selection, training methods, etc., and collect additional data. Multiple rounds of iterations can be performed to continuously improve the model performance until the design indicators are met.

[0107] For example, assume that the distribution out-generalization data includes: (1) high-speed conditions: vehicle speed is between 120 and 160 kilometers per hour; (2) large-angle conditions: steering wheel angle is between 400 and 500 degrees; (3) heavy-load conditions: vertical load is between 1500 and 2000 kilograms; (4) extreme conditions: vehicle speed is 160 kilometers per hour, steering wheel angle is 500 degrees, and vertical load is 2000 kilograms.

[0108] The prediction effect of the steering wheel torque prediction model is evaluated on the out-of-distribution generalization data (1) to (4), and the following evaluation results can be obtained:

[0109] (1) High-speed working condition: MAE = 0.4 N·m, MSE = 0.3 (N·m) 2 ;

[0110] (2) Large turning angle condition: MAE = 0.5 N·m, MSE = 0.4 (N·m) 2 ;

[0111] (3) Heavy load condition: MAE = 0.6 N·m, MSE = 0.5 (N·m) 2 ;

[0112] (4) Extreme working conditions: MAE = 0.8 N·m, MSE = 0.7 (N·m) 2 .

[0113] It can be seen that under the 00D condition, the prediction error of the steering wheel torque prediction model has increased, but it is still within an acceptable range overall. The error under extreme conditions is the largest and needs to be focused on and optimized. For example, if after a period of testing, it is found that the prediction error of the model under certain special conditions (such as rainy days, muddy roads, etc.) is large, the data of these conditions can be collected in a targeted manner and the model can be retrained.

[0114] For example, the following strategy can be used to iteratively optimize the steering wheel torque prediction model:

[0115] (1) Data enhancement: The existing data is disturbed by adding Gaussian noise, random scaling, rotation, etc., and the physical model is used to generate more extreme operating condition data, such as high-speed sharp turns and emergency braking. At the same time, more real vehicle data is collected, especially data under severe weather and complex road conditions, to enrich the diversity of the training set and improve the generalization ability of the model.

[0116] (2) Feature Engineering: Review the features used in the model, introduce high-discrimination features related to steering characteristics such as vehicle yaw rate and tire slip angle, optimize model input, and improve its adaptability to OOD data.

[0117] (3) Loss function optimization: Design a complex loss function. On the basis of mean square error loss, add the weighted error term of OOD data to increase the model's attention to OOD data. Introduce robust loss functions such as Huber loss and Tukey loss to reduce the impact of abnormal data points. Design an adaptive loss function to dynamically adjust the weight according to the difficulty of the data point, and guide the model to better learn the characteristics of OOD data.

[0118] As an example, suppose we evaluate the generalization ability of the model on the OOD test set and find that the model's prediction error is large and cannot meet the design requirements. Suppose the performance of the original model on the OOD test set is:

[0119] (1) High-speed operation: MAE = 0.8 N·m, MSE = 0.7 (N·m) 2 ;

[0120] (2) Large turning angle condition: MAE = 1.0 N·m, MSE = 0.9 (N·m) 2 ;

[0121] (3) Heavy load condition: MAE = 1.2 N·m, MSE = 1.1 (N·m) 2 ;

[0122] (4) Extreme working conditions: MAE = 1.5 N·m, MSE = 1.4 (N·m) 2 .

[0123] This shows that the current model has poor generalization ability under OOD conditions and needs to be optimized.

[0124] Assume that after three rounds of iterations, a model that meets the design requirements is obtained, and its performance on the OOD test set is as follows:

[0125] (1) High-speed working condition: MAE = 0.3 N·m, MSE = 0.2 (N·m) 2 ;

[0126] (2) Large turning angle condition: MAE = 0.4 N·m, MSE = 0.3 (N·m) 2 ;

[0127] (3) Heavy load condition: MAE = 0.5 N·m, MSE = 0.4 (N·m) 2 ;

[0128] (4) Extreme working conditions: MAE = 0.6 N·m, MSE = 0.5 (N·m) 2 .

[0129] The optimized model can give relatively accurate predictions under various OOD conditions, and its generalization ability has been significantly improved, meeting the design requirements.

[0130] The steering wheel torque prediction model in this application has undergone multiple rounds of iterative optimization, and its performance and generalization ability have been continuously improved, ultimately meeting the design indicators. It can stably and reliably predict the steering wheel torque under various complex working conditions, providing strong support for the control of the steering system.

[0131] It is understandable that in practical applications, it is necessary to flexibly select and combine strategies such as data enhancement, feature engineering, model structure adjustment, and loss function optimization according to specific problems and data characteristics, and continuously monitor model performance and adjust the optimization direction in a timely manner. Through continuous iterative optimization, a robust and reliable model is finally obtained to ensure that the wire-controlled steering system can stably and reliably predict the steering wheel torque under various complex working conditions, providing strong support for the control of the steering system. In short, model optimization is a process of continuous iteration and dynamic adjustment. It is necessary to comprehensively consider the generalization ability of the model, computational efficiency, and other factors, and select the optimal combination of optimization strategies according to actual needs and constraints. Through continuous optimization and improvement, a high-performance model that meets the design requirements and can operate stably in a production environment is finally obtained, providing a reliable and efficient wire-controlled steering system torque feedback prediction verification function for the wire-controlled steering system.

[0132] Step S360: When the iteratively optimized steering wheel torque prediction model meets the design requirements, the current steering wheel torque prediction model is obtained.

[0133] When the iteratively optimized steering wheel torque prediction model meets the design requirements, the model performance is good. At this time, the current steering wheel torque prediction model can be obtained and deployed to a local or cloud control unit to replace the original verification model, thereby using the steering wheel torque prediction model to predict the steering wheel torque.

[0134] Step S370: using the steering wheel torque prediction model to predict the steering wheel torque.

[0135] After using the steering wheel torque prediction model to predict the steering wheel torque, the actual performance of the steering wheel torque prediction model can be monitored, feedback data can be collected, and preparations can be made for the next round of iterations. Through this continuous iteration approach, the model can always maintain high performance and generalization capabilities and adapt to the ever-changing actual working conditions.

[0136] For a detailed description of step S370 , please refer to step S130 .

[0137] Compared with step S210 to step S250, step S310 to step S370 also have the following additional technical effects: through continuous iterative optimization, the steering wheel torque prediction model can always maintain high performance and generalization ability, so as to adapt to the ever-changing actual working conditions and improve the robustness and generalization ability of the steering wheel torque prediction model. This application establishes a SBW system road feel feedback performance evaluation and verification method based on cloud data, which can more comprehensively and systematically examine the actual performance of the SBW system under various complex working conditions.

[0138] In some embodiments, after the steering wheel torque prediction model is used to predict the steering wheel torque, that is, after the steering wheel torque prediction model is deployed as a new version in the server, experimental verification can also be performed to comprehensively evaluate the actual performance of the steering wheel torque prediction model. Specifically, a bench test can be performed first, and then a real vehicle test can be performed on the basis of the bench test. The bench test includes: a preliminary test can be first performed on the bench of the wire control steering system to verify the effectiveness of the steering wheel torque prediction model. In the test, the steering wheel torque prediction model is trained and tested using data under different working conditions to evaluate its prediction accuracy and real-time performance. The real vehicle test includes: by collecting real vehicle data on actual roads, evaluating the prediction effect of the model in a real environment, thereby focusing on the performance of the model under extreme working conditions and abnormal conditions, and verifying its generalization ability and robustness. By combining bench tests and real vehicle tests, the actual performance of the SBW system is comprehensively evaluated, providing a basis for its further optimization and improvement, and promoting the development and application of SBW technology.

[0139] See also Figure 8 , Figure 8A structural schematic diagram of a steering wheel torque prediction device provided in an embodiment of the present application is shown. The steering wheel torque prediction device 200 can be applied to a server. The steering wheel torque prediction device 200 may include: a data acquisition module 210, a model training module 220 and a torque prediction module 230. Among them: the data acquisition module 210 is used to obtain real vehicle data, and the real vehicle data includes the steering wheel angle, angular velocity, vehicle speed, lateral acceleration, vertical load and steering wheel torque of the vehicle. The model training module 220 is used to determine the steering wheel torque prediction model based on the real vehicle data. The torque prediction module 230 is used to predict the steering wheel torque using the steering wheel torque prediction model.

[0140] In some embodiments, the model training module 220 is also used to add multiple penalty terms to the steering wheel torque prediction model to obtain an objective function, wherein the multiple penalty terms include error penalty, extreme value penalty and smoothness penalty for out-of-distribution generalized data; and the generalization ability of the steering wheel torque prediction model is evaluated based on the actual vehicle data and the objective function.

[0141] In some embodiments, the model training module 220 is also used to perform distribution shift on the real vehicle data to obtain out-of-distribution generalized data; input the out-of-distribution generalized data into the objective function to obtain the steering wheel torque predicted by the objective function; and evaluate the generalization ability of the steering wheel torque prediction model based on the error between the steering wheel torque predicted by the objective function and the steering wheel torque collected from the real vehicle.

[0142] In some embodiments, the model training module 220 is also used to apply a random offset to the real vehicle data to obtain mean offset data; apply a random scaling factor to the real vehicle data to obtain variance offset data; perform extreme value amplification or reduction on the real vehicle data to obtain extreme value offset data; combine the mean offset data, the variance offset data, and the extreme value offset data to obtain distribution out-of-distribution generalization data.

[0143] In some embodiments, the model training module 220 is also used to iteratively optimize the steering wheel torque prediction model when the evaluation result does not meet the design requirements; when the iteratively optimized steering wheel torque prediction model meets the design requirements, obtain the current steering wheel torque prediction model.

[0144] In some embodiments, the model training module 220 is further used to construct a steering wheel torque prediction model as shown in the following expression:

[0145] T=k0+k1·θ+k2·ω+k3·v+k4·a+k5·F+k6·θ 2 +k7·ω 2 +k8·v 2 +k9·a2 +k 10 ·F 2 +k 11 ·θ·ω+k 12 ·θ·v+k 13 ·θ·a+k 14 ·θ·F+k 15 ·ω·v+k 16 ·ω·a+k 17 ·ω·F+k 18 ·v·a+k 19 ·v·F+k 20 ·a·F

[0146] Where T is the steering wheel torque, θ is the steering wheel angle, ω is the angular velocity, v is the vehicle speed, a is the lateral acceleration, F is the vertical load, k0 to k 20 are model parameters to be fitted; according to the actual vehicle data, the model parameters in the steering wheel torque prediction model are fitted to obtain the steering wheel torque prediction model.

[0147] In some embodiments, the model training module 220 is further used to fit the model parameters in the steering wheel torque prediction model according to the real vehicle data by using the least squares method to obtain the steering wheel torque prediction model.

[0148] Those skilled in the art can clearly understand that the above device provided in the embodiment of the present application can implement the method provided in the embodiment of the present application. The specific working process of the above-described device and module can refer to the corresponding process of the method in the embodiment of the present application, which will not be repeated here.

[0149] In the embodiments provided in the present application, the coupling, direct coupling or communication connection between the modules shown or discussed may be indirect coupling or communication coupling through some interfaces, devices or modules, and may be electrical, mechanical or other forms, and the embodiments of the present application do not impose specific limitations on this.

[0150] In addition, each functional module in the embodiment of the present application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The above integrated modules can be implemented in the form of hardware or in the form of software functional modules.

[0151] See also Fig. 9 , Fig. 9 The server 300 may include a memory 310 and a processor 320 , wherein the memory 310 stores an application program, and the application program is configured to execute the method provided in the embodiment of the present application when called by the processor 320 .

[0152] The processor 320 may include one or more processing cores. The processor 320 uses various interfaces and lines to connect various parts of the entire server 300, and is used to run or execute instructions, programs, code sets or instruction sets stored in the memory 310, and call to run or execute data stored in the memory 310, perform various functions of the server 300 and process data.

[0153] The processor 320 can be implemented in at least one of the following hardware forms: digital signal processing (DSP), field programmable gate array (FPGA), and programmable logic array (PLA). The processor 320 can integrate one or a combination of a central processing unit (CPU), a graphics processing unit (GPU), and a modem. Among them, the CPU mainly processes the operating system, user interface, and application programs; the GPU is responsible for rendering and drawing display content; and the modem is used to process wireless communications. It is understandable that the above-mentioned modem may not be integrated into the processor 320, but may be implemented separately through a communication chip.

[0154] The memory 310 may include a random access memory (RAM) or a read-only memory (ROM). The memory 310 may be used to store instructions, programs, codes, code sets or instruction sets. The memory 310 may include a program storage area and a data storage area. The program storage area may store instructions for implementing an operating system, instructions for implementing at least one function, instructions for implementing the above-mentioned various method embodiments, etc. The data storage area may store data created by the server 300 during use, etc.

[0155] An embodiment of the present application also provides a computer-readable storage medium, on which a program code is stored. The program code is configured to execute the method provided in the embodiment of the present application when called by a processor.

[0156] The computer-readable storage medium may be an electronic memory such as a flash memory, an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a hard disk or a ROM.

[0157] In some embodiments, the computer readable storage medium includes a non-volatile computer readable medium (Non-Transitory Computer-Readable Storage Medium, referred to as Non-TCRSM). The computer readable storage medium has a storage space for the program code that executes any method step in the above method. These program codes can be read from or written into one or more computer program products. The program code can be compressed in an appropriate form.

[0158] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A steering wheel torque prediction method, characterized in that: include: Acquiring real vehicle data, wherein the real vehicle data includes a steering wheel angle, angular velocity, vehicle speed, lateral acceleration, vertical load, and steering wheel torque of the vehicle; Determining a steering wheel torque prediction model according to the actual vehicle data; The steering wheel torque prediction model is used to predict the steering wheel torque.

2. The method according to claim 1, characterized in that Before predicting the steering wheel torque using the steering wheel torque prediction model, the method further includes: Adding multiple penalty terms to the steering wheel torque prediction model to obtain an objective function, wherein the multiple penalty terms include error penalty for out-of-distribution generalized data, extreme value penalty, and smoothness penalty; The generalization ability of the steering wheel torque prediction model is evaluated based on the actual vehicle data and the objective function.

3. The method according to claim 2, characterized in that The step of evaluating the generalization ability of the steering wheel torque prediction model according to the real vehicle data and the objective function includes: Performing distribution shift on the real vehicle data to obtain out-of-distribution generalized data; Inputting the out-of-distribution generalized data into the objective function to obtain the steering wheel torque predicted by the objective function; The generalization ability of the steering wheel torque prediction model is evaluated based on the error between the steering wheel torque predicted by the objective function and the steering wheel torque collected from the actual vehicle.

4. The method according to claim 3, characterized in that The step of performing distribution shift on the real vehicle data to obtain distribution out-of-distribution generalized data includes: Applying a random offset to the real vehicle data to obtain mean offset data; Applying a random scaling factor to the real vehicle data to obtain variance offset data; Performing extreme value magnification or reduction on the real vehicle data to obtain extreme value offset data; The mean shift data, the variance shift data, and the extreme value shift data are combined to obtain distribution out-of-distribution generalization data.

5. The method according to any one of claims 2 to 4, characterized in that: After evaluating the generalization ability of the steering wheel torque prediction model according to the real vehicle data and the objective function, the method further includes: When the evaluation result does not meet the design requirements, iteratively optimizing the steering wheel torque prediction model; When the iteratively optimized steering wheel torque prediction model meets the design requirements, the current steering wheel torque prediction model is obtained.

6. The method according to claim 1, characterized in that Determining the steering wheel torque prediction model according to the actual vehicle data includes: Construct a steering wheel torque prediction model as shown in the following expression: T=k0+k1·θ+k2·ω+k3·v+k4·a+k5·F+k6·θ 2 +k7·ω 2 +k8·v 2 +k9·a 2 +k 10 ·F 2 +k 11 ·θ·ω+k 12 ·θ·v+k 13 ·θ·a+k 14 ·θ·F+k 15 ·ω·v+k 16 ·ω·a+k 17 ·ω·F+k 18 ·v·a+k 19 ·v·F+k 20 ·a·F Where, T is the steering wheel torque, θ is the steering wheel angle, ω is the angular velocity, v is the vehicle speed, a is the lateral acceleration, F is the vertical load, k0 to k 20 are the model parameters to be fitted; P1CN-GQJT240221 and 2024P0905CN1 fit the model parameters in the steering wheel torque prediction model according to the actual vehicle data to obtain the steering wheel torque prediction model.

7. The method according to claim 6, characterized in that The step of fitting the model parameters in the steering wheel torque prediction model according to the actual vehicle data to obtain the steering wheel torque prediction model comprises: According to the actual vehicle data, the least square method is used to fit the model parameters in the steering wheel torque prediction model to obtain the steering wheel torque prediction model.

8. A steering wheel torque prediction device, characterized in that: include: A data acquisition module is used to obtain real vehicle data, wherein the real vehicle data includes the vehicle's steering wheel angle, angular velocity, vehicle speed, lateral acceleration, vertical load, and steering wheel torque; A model training module, used to determine a steering wheel torque prediction model based on the real vehicle data; The torque prediction module is used to predict the steering wheel torque using the steering wheel torque prediction model.

9. A server, characterized in that: include: A memory and a processor, wherein an application is stored in the memory, and the application is configured to cause the processor to execute the method according to any one of claims 1 to 7 when called by the processor.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program codes, and the program codes are configured to, when called by a processor, cause the processor to execute the method according to any one of claims 1 to 7.

Images