A method for constructing a steering rack force prediction model and related devices

Abstract

The application discloses a steering rack force prediction model construction method and related device, a sample set is constructed by acquiring vehicle parameters and corresponding real rack force, and an intermediate variable is introduced between the vehicle parameters and the rack force according to a theoretical formula which has a function relationship with the vehicle parameters and the rack force, part of the determined relationship between the vehicle parameters and the rack force is captured in the complex relationship between the vehicle parameters and the rack force, the influence of the complex changes between the vehicle parameters on the corresponding rack force is helped to be explained, so that a rack force regression prediction model with higher prediction accuracy can be constructed by a supervised method, the accurate prediction of the rack force required by vehicle steering under different driving states is realized, the host computer can limit the expected steering speed of the tire according to the actual required rack force, the situation that the tire steering speed cannot follow the host computer instruction is avoided, the stability and safety of the vehicle are improved, and a large amount of cost and expense is saved.

Description

Technical Field

[0001] This application relates to the field of vehicle steering technology, specifically to a method for constructing a steering rack force prediction model and related apparatus. Background Technology

[0002] During the vehicle's tire steering process, the host computer is responsible for issuing control commands. The steering system follows the requested angle and angle slope sent by the host computer. Since the motor power of the steering system is constant, when the actual required rack force is close to or exceeds the maximum output torque of the motor, it will cause the output rate of the steering system to decrease, resulting in the tire steering speed being lower than the expected value sent by the host computer, and a delayed following phenomenon will occur. The host computer needs to combine the rack force actually required for tire steering with the capability of the steering system to limit the expected steering speed so that the tire steering speed can follow the commands of the host computer.

[0003] In related technologies, rack force can be estimated in two ways: one is to directly measure the rack force by installing sensors on the vehicle, and the other is to estimate the rack force using a theoretical model.

[0004] However, the relevant technologies struggle to accurately predict rack forces under different driving conditions at a low cost. Summary of the Invention

[0005] In view of this, this application provides a method and related apparatus for constructing a steering rack force prediction model. Based on a theoretical formula that can represent the functional relationship between some vehicle parameters and the actual rack force, relevant variables are introduced into the training samples to help the model capture and interpret the laws and relationships between vehicle parameters and rack force, thereby improving the accuracy of the prediction model.

[0006] To solve the above problems, the technical solution provided in this application is as follows:

[0007] On the one hand, this application provides a method for constructing a steering rack force prediction model, including:

[0008] Obtain a first training sample set. The model input of the first training sample set includes multiple sets of vehicle parameters. The sample label of the first training sample set includes the rack force corresponding to each set of vehicle parameters. Each set of vehicle parameters is used to identify the driving state of the vehicle. The rack force is used to control the tire steering action of the vehicle.

[0009] Obtain the theoretical formula, which is used to identify the functional relationship between each set of vehicle parameters and the corresponding rack force;

[0010] Based on each set of vehicle parameters and the theoretical formula, the intermediate variables between each set of vehicle parameters and the corresponding rack force are determined, and the intermediate variables are added as model inputs to the first training sample set to obtain the second training sample set. The intermediate variables are the relevant quantities of the theoretical formula.

[0011] Based on the second training sample set, a rack force prediction model is constructed using supervised training. The rack force prediction model is used to predict the corresponding rack force based on vehicle parameters under different driving conditions.

[0012] In one possible implementation, each set of vehicle parameters includes:

[0013] Vehicle speed, tire vertical load, road surface friction coefficient, steering wheel angle, tire angle, and tire angular velocity.

[0014] In one possible implementation, the theoretical formula includes:

[0015] The theoretical formulas for steering lever ratio, tire resistance torque, and gravity return torque are provided. The steering lever ratio is used to identify the relationship between tire angle and rack travel. The product of gravity return torque and tire resistance torque with rack travel is used to identify rack force.

[0016] In one possible implementation, the theoretical formula for the steering lever ratio is expressed by the following formula:

[0017]

[0018] Where b is the steering lever ratio, δ is the tire angle, and S is the rack travel.

[0019] In one possible implementation, the theoretical formula for the gravity-correcting torque is expressed by the following formula:

[0020]

[0021] Among them, T Fz The gravity-correcting torque is given by α = ε + γ, where θ is the wheel rotation angle around the kingpin, ε is the kingpin inclination angle, σ is the kingpin caster angle, and γ is the wheel camber angle. py F is the lateral offset of the kingpin at the wheel center. z The vertical load on the tire is denoted as .

[0022] In one possible implementation, the theoretical formula for the tire drag torque is expressed by the following formula:

[0023]

[0024] Among them, T MzThe tire drag torque is μ, the road surface friction coefficient is F. z ρ is the vertical load on the tire, and p is the tire pressure.

[0025] In one possible implementation, the intermediate variables between each set of vehicle parameters and the corresponding rack force include:

[0026] The steering lever ratio, the tire drag torque, and the gravity return torque for each set of vehicle parameters.

[0027] In one possible implementation, the construction of the rack force prediction model using supervised training includes:

[0028] The rack force prediction model was constructed using the CART regression algorithm.

[0029] In another aspect, this application provides a device for constructing a steering rack force prediction model, including an acquisition unit, a determination unit, and a construction unit:

[0030] The acquisition unit is used to acquire a first training sample set. The model input of the first training sample set includes multiple sets of vehicle parameters. The sample label of the first training sample set includes the rack force corresponding to each set of vehicle parameters. Each set of vehicle parameters is used to identify the driving state of the vehicle. The rack force is used to control the tire steering action of the vehicle.

[0031] The acquisition unit is also used to acquire theoretical formulas, which are used to identify the functional relationships between each set of vehicle parameters and the corresponding rack force.

[0032] The determining unit is used to determine the intermediate variable between each set of vehicle parameters and the corresponding rack force according to each set of vehicle parameters and the theoretical formula, and add the intermediate variable as model input to the first training sample set to obtain the second training sample set. The intermediate variable is the relevant quantity of the theoretical formula.

[0033] The construction unit is used to construct a rack force prediction model using supervised training based on the second training sample set. The rack force prediction model is used to predict the corresponding rack force based on the vehicle parameters under different driving conditions.

[0034] In one possible implementation, each set of vehicle parameters includes:

[0035] Vehicle speed, tire vertical load, road surface friction coefficient, steering wheel angle, tire angle, and tire angular velocity.

[0036] In one possible implementation, the theoretical formula includes:

[0037] The theoretical formulas for steering lever ratio, tire resistance torque, and gravity return torque are provided. The steering lever ratio is used to identify the relationship between tire angle and rack travel. The product of gravity return torque and tire resistance torque with rack travel is used to identify rack force.

[0038] In one possible implementation, the theoretical formula for the steering lever ratio is expressed by the following formula:

[0039]

[0040] Where b is the steering lever ratio, δ is the tire angle, and S is the rack travel.

[0041] In one possible implementation, the theoretical formula for the gravity-correcting torque is expressed by the following formula:

[0042]

[0043] Among them, T Fz The gravity-correcting torque is given by α = ε + γ, where θ is the wheel rotation angle around the kingpin, ε is the kingpin inclination angle, σ is the kingpin caster angle, and γ is the wheel camber angle. py F is the lateral offset of the kingpin at the wheel center. z The vertical load on the tire is denoted as .

[0044] In one possible implementation, the theoretical formula for the tire drag torque is expressed by the following formula:

[0045]

[0046] Among them, T Mz The tire drag torque is μ, the road surface friction coefficient is F. z ρ is the vertical load on the tire, and p is the tire pressure.

[0047] In one possible implementation, the intermediate variables between each set of vehicle parameters and the corresponding rack force include:

[0048] The steering lever ratio, the tire drag torque, and the gravity return torque for each set of vehicle parameters.

[0049] In one possible implementation, the building block is used to construct the rack force prediction model using the CART regression algorithm.

[0050] In another aspect, this application provides a computer device, including a processor and a memory:

[0051] The memory is used to store computer programs;

[0052] The processor is used to execute the steering rack force prediction model construction method according to the computer program.

[0053] As can be seen from the above technical solution, the computer equipment acquires vehicle parameters under different driving conditions and the rack force used to control tire steering under the corresponding conditions. The vehicle parameters are used as input and the rack force is used as a label to form a first training sample set. Then, a theoretical formula that has a functional relationship with both the vehicle parameters and the rack force is obtained. Based on the vehicle parameters under different driving conditions and the theoretical formula, intermediate variables that can describe the definite relationship between some vehicle parameters and rack force are introduced into the input of the first training sample set to form a second training sample set. Then, a rack force prediction model is constructed using a supervised training method based on the second training sample set to predict the rack force required by the vehicle under different driving conditions.

[0054] This technical solution introduces intermediate variables into the training sample set using theoretical formulas that describe the complex relationship between vehicle driving parameters and rack force. This allows for the acquisition of functional relationships between some vehicle parameters and rack force within the complex and unknown relationship between them. This helps to capture and explain the impact of complex changes in vehicle parameters on rack force, constructing a rack force regression prediction model with higher accuracy. This enables precise prediction of the rack force required for vehicle steering under different driving conditions. By feeding the predicted rack force and steering system capabilities back to the host computer, the desired steering speed can be adjusted, preventing situations where tire steering speed cannot keep up, thus improving vehicle stability and safety. Furthermore, since vehicle parameters are information that can be directly obtained without adding sensors or other equipment, using a prediction model to predict rack force under different driving conditions can save significant costs and expenses. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1 A flowchart illustrating a method for constructing a steering rack force prediction model provided in an embodiment of this application;

[0057] Figure 2 A schematic diagram of a "tire angle - rack travel" curve provided for an embodiment of this application;

[0058] Figure 3A comparison chart of the actual and predicted values ​​of rack force provided for embodiments of this application;

[0059] Figure 4 A comparison chart of the actual and predicted values ​​of rack force provided for another embodiment of this application;

[0060] Figure 5 A schematic diagram illustrating a rear wheel steering rack force prediction process provided in an embodiment of this application;

[0061] Figure 6 This is a schematic diagram of a steering rack force prediction model construction device provided in an embodiment of this application. Detailed Implementation

[0062] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0063] As described in the background section, the steering system follows the request angle and angle slope sent by the host computer. Since the motor power of the steering system is constant, when the actual required rack force is close to or exceeds the maximum output torque of the motor, the output rate of the steering system will decrease, causing the tire steering speed to be lower than the expected value sent by the host computer, resulting in a delayed following phenomenon. Current methods are difficult to achieve accurate prediction of rack force at low cost, and therefore cannot limit the expected steering speed based on the actual required rack force to avoid the situation where the tire steering speed cannot follow the instructions of the host computer.

[0064] This application provides a method for constructing a steering rack force prediction model. During the model training process, intermediate variables related to vehicle parameters and rack force are introduced. By obtaining the functional relationship between some parameters and rack force from the unknown relationship between vehicle parameters and rack force under different driving conditions, it helps to capture and interpret the influence of complex changes in vehicle parameters on the corresponding rack force. This constructs a rack force regression prediction model with higher prediction accuracy, thereby avoiding the situation where the tire steering speed cannot keep up when the vehicle uses this model to predict the rack force required under different driving conditions.

[0065] The method provided in this application is mainly applied to vehicles using steer-by-wire technology, that is, vehicles that transmit steering commands through electronic signals. These vehicles can be engineering vehicles used in the engineering field, such as excavators and bulldozers, or vehicles used for daily travel, such as cars, electric vehicles, and sports utility vehicles. This application does not limit the use of the vehicle.

[0066] The solutions provided in this application involve various technologies such as vehicle steering technology and machine learning. A predictive model of the rack force required for steering is constructed using machine learning methods. This predictive model is then encapsulated into a software module and deployed on different vehicles. This allows the predictive model to accurately predict the rack force required for steering under different driving conditions. As a result, when steering using vehicle steering technology, the vehicle's steering system can achieve more intelligent control based on the real-time predicted rack force, thereby improving the vehicle's safety and stability.

[0067] See Figure 1 The diagram shown is a flowchart illustrating a method for constructing a steering rack force prediction model according to an embodiment of this application. In this embodiment, it can be executed by a computer device.

[0068] S101: Obtain the first training sample set. The model input of the first training sample set includes multiple sets of vehicle parameters, and the sample labels of the first training sample set include the rack force corresponding to each set of vehicle parameters.

[0069] Each set of vehicle parameters is used to identify the vehicle's driving status, and the rack force is used to control the vehicle's tire steering.

[0070] Vehicle parameters refer to relevant information about the vehicle under different driving conditions, which can be obtained in real time through the ECU. Rack force is the force acting on the steering rack during vehicle steering, which needs to be obtained by installing sensors on the vehicle.

[0071] Vehicle parameters include both parameters inherent to the vehicle itself and parameters generated by external environmental factors. Vehicle-inherent parameters include tire pressure, vehicle speed, and mileage. Parameters generated by external environmental factors include temperature, tire vertical load, road surface friction coefficient, tire angle, and tire angular velocity.

[0072] When constructing a rack force prediction model for the same type of vehicle, different driving conditions should be simulated using the same type of vehicle. By artificially altering some vehicle parameters, richer and more diverse sample data can be obtained to ensure the authenticity and reliability of the samples. For example, the vehicle's driving scenarios need to cover lane changes, right-angle turns, circular maneuvers, reversing, and stationary starting and turning. Under different driving scenarios, the steering wheel angle, tire angle, tire angular velocity, and vehicle speed will all change. In this process, different counterweights can be used to change the vertical load on the tires, and the road friction coefficient can be changed by driving on different road surfaces, thus encompassing information on complex vehicle parameters in actual driving scenarios.

[0073] The computer device acquires a large number of vehicle parameters and corresponding rack forces collected under different driving scenarios. The vehicle parameters are used as model inputs, and the corresponding rack forces are used as sample labels to obtain the first training sample set.

[0074] S102: Obtain the theoretical formula.

[0075] The theoretical formula is used to identify the functional relationship between each set of vehicle parameters and the corresponding rack force.

[0076] Functional relationships represent definite relationships that can be derived from mathematical formulas. Under different driving conditions, vehicle parameters constantly change, and the complex relationships between these parameters make it difficult for models to discover the patterns and connections between vehicle parameters and rack forces. Under relatively stable driving conditions, i.e., when some vehicle parameters remain constant, rack forces can be estimated using theoretical formulas. While theoretical formulas alone provide estimates of rack forces under ideal conditions, they can reveal functional relationships between some vehicle parameters and rack forces, helping to explain the patterns and connections between other vehicle parameters and rack forces.

[0077] As an example, rack force is the ratio of torque to lever arm. Torque includes steering assist, friction, and tire reaction force. The tire reaction force is functionally related to vehicle parameters such as tire vertical load and road surface friction coefficient. Computer equipment can obtain theoretical formulas related to the aforementioned torque to establish a definite relationship between some vehicle parameters and rack force.

[0078] Computer equipment can acquire theoretical formulas, decompose the relationship between some vehicle parameters and their corresponding rack forces into a functional relationship that can be expressed by mathematical formulas, help the model establish a definite relationship between some vehicle parameters and rack forces, thereby discovering the regularity between vehicle parameters and their corresponding rack forces and improving the interpretability of the model.

[0079] S103: Based on each set of vehicle parameters and theoretical formulas, determine the intermediate variables between each set of vehicle parameters and the corresponding rack force, and add the intermediate variables as model inputs to the first training sample set to obtain the second training sample set.

[0080] The intermediate variables are the relevant quantities in the theoretical formula. For each set of vehicle parameters, the computer equipment uses the theoretical formula to obtain the intermediate variables between the vehicle parameters and the corresponding rack force. These intermediate variables are then added as input to the first training sample set to obtain a second training sample set with more inputs.

[0081] As an example, the theoretical formula includes a formula for calculating tire reaction force based on the vertical load on the tire. The computer device then uses the acquired theoretical formula to calculate the intermediate variable corresponding to each set of vehicle parameters, namely the tire reaction force. The computer device then uses the tire reaction force as input to the corresponding samples to obtain the second training sample set.

[0082] The second training sample set obtained by the computer equipment introduces intermediate variables that can represent the functional relationship between the input and the label compared to the first training sample set. This helps the model explain the impact of complex changes in vehicle parameters on the calculation of rack force, thus obtaining a rack force prediction model with higher accuracy.

[0083] S104: Based on the second training sample set, construct a rack force prediction model using supervised training.

[0084] The rack force prediction model is used to predict the corresponding rack force based on vehicle parameters under different driving conditions. The computer takes the vehicle parameters and intermediate variables as model inputs, uses the corresponding rack force as a label, and trains the model using a suitable supervised algorithm to obtain the rack force prediction model. During the vehicle's tire steering motion, the required rack force is a continuously changing value; therefore, a regression algorithm is chosen for the supervised algorithm, such as linear regression, support vector regression, random forest regression, or decision tree regression.

[0085] As an example, the computer device uses the vehicle parameters and intermediate variables from the second training sample set as model input, the corresponding rack force as label, selects the decision tree regression algorithm, and trains the model by defining the loss function, optimization algorithm, and other model-related parameters. During the training process, the model parameters are continuously adjusted according to the optimization algorithm, so that the value of the loss function gradually decreases until the loss function reaches the preset value. Thus, a well-trained rack force prediction model is obtained.

[0086] After obtaining the rack force prediction model, the model is packaged into software and deployed in the vehicle's system. By obtaining the vehicle parameters required by the model, the tire steering rack force of the vehicle under different driving conditions can be predicted in real time.

[0087] Therefore, by introducing intermediate variables into the training sample set using theoretical formulas that describe the definite relationship between certain vehicle parameters and rack force, computer equipment establishes a functional relationship between certain vehicle parameters and rack force. This helps capture and interpret the impact of complex changes in vehicle parameters on rack force, constructing a rack force prediction model with higher accuracy. This model can then be used on the vehicle to accurately predict the rack force required for steering under different driving conditions. The host computer can then adjust the desired steering speed of the tires, avoiding a situation where the estimated rack force is higher than the actual value, leading to a decrease in the steering system's output rate and causing the tire steering speed to lag behind the desired steering speed, thus improving vehicle stability and safety. Furthermore, using the prediction model to accurately predict rack force under different driving conditions allows for rack force calculation without adding sensors or other equipment, saving significant costs and achieving a balance between cost and accuracy.

[0088] To reduce the complexity of model training, important parameters that are relevant to the calculation of rack force were selected. In one possible implementation, each set of vehicle parameters includes: vehicle speed, tire vertical load, road friction coefficient, steering wheel angle, tire angle, and tire angular velocity.

[0089] Vehicle speed changes affect the vehicle's dynamic load and steering response characteristics, thus affecting the rack force required for steering. Tire vertical load refers to the vertical component of the force exerted on the tire by the road surface. The road friction coefficient is an important indicator for measuring the road's anti-skid performance. Tire angle and tire angular velocity are parameters closely related to tire rotation. Steering wheel angle directly affects tire angle. Therefore, all of the above vehicle parameters are related to the rack force required for tire steering. In this embodiment, the computer device uses the above vehicle parameters as model input for the first training sample set.

[0090] As an example, for a rear-wheel drive vehicle, when calculating the rear wheel steering rack force, the tire vertical load is the rear axle load, the tire angle is the rear wheel angle, and the tire angular velocity is the rear wheel angular velocity. Table 1 shows the model input and labels for the first training sample set under this example.

[0091] Table 1

[0092]

[0093] The units of model input and labels should remain consistent across different samples to facilitate model training.

[0094] By using vehicle parameters that are relevant to tire steering actions as model inputs, interference from irrelevant parameters can be avoided during model training, reducing model complexity and further improving model prediction efficiency.

[0095] During model training, in order to better learn the relationship between the input and the label, a theoretical formula that has a functional relationship with both the input and the rack force is selected. In one possible implementation, the theoretical formulas include: the theoretical formulas for the steering lever ratio, tire resistance torque, and gravity return torque.

[0096] The steering lever ratio is used to indicate the relationship between the tire angle and the rack travel. The product of the gravity return torque and the tire resistance torque with the rack travel is used to indicate the rack force.

[0097] Steering lever ratio is a parameter describing the relationship between tire angle and rack travel. Gravity return torque is the torque that returns the wheel to a straight position due to gravity during steering. Tire drag torque is the torque that returns the wheel to a straight position due to the lateral friction between the tire and the ground.

[0098] The gravity-aligning torque and tire steering torque are closely related to the vehicle's driving state and have a definite functional relationship with some vehicle parameters. They can be calculated using these parameters. Furthermore, during vehicle steering, the rack force can be estimated by multiplying the gravity-aligning torque, tire steering torque, and rack travel. In other words, the gravity-aligning torque, tire steering torque, and rack travel have a many-to-one functional relationship with the rack force. Therefore, the theoretical formulas for the gravity-aligning torque, tire steering torque, and steering lever ratio can establish a functional relationship between some vehicle parameters and the rack force. In this embodiment, the theoretical formulas for the steering lever ratio, gravity-aligning torque, and tire drag torque are chosen to help the model capture and explain the relationship between vehicle parameters and rack force.

[0099] Furthermore, the rack force required to overcome during tire steering includes not only the gravity-aligning torque and tire steering torque, but also inertial torque and suspension system torque. However, the gravity-aligning torque and tire steering torque are the two most significant torques that need to be overcome during vehicle steering. The rack force obtained by considering the gravity-aligning torque and tire drag torque meets the accuracy requirements of most application scenarios. Therefore, to simplify the rack force prediction model, only the effects of the gravity-aligning torque and tire drag torque are considered when estimating the rack force.

[0100] The steering leverage ratio is the ratio of the tire angle to the rack travel. In one possible implementation, the theoretical formula for the steering leverage ratio is expressed as follows:

[0101]

[0102] Where b is the steering lever ratio, δ is the tire angle, and S is the rack travel.

[0103] like Figure 2 The diagram shown is a schematic representation of a "tire angle - rack travel" curve provided in an embodiment of this application. Due to the geometric design of the steering system, the steering lever ratios of the left and right wheels are typically different. Using the above formula, the rack travel of each wheel can be obtained based on the steering lever ratio and the change in tire angle, and the average value is taken as the rack travel required to estimate the rack force when the front or rear wheels are steering. In actual driving, the steering lever ratio can be dynamically adjusted according to driving conditions or driving mode.

[0104] The gravity-aligning torque is one of the main torques that causes a tire to return to a straight-line driving position due to gravity when turning. Calculating rack force during vehicle turning requires first considering the influence of the gravity-aligning torque. In one possible implementation, the theoretical formula for the gravity-aligning torque is expressed as follows:

[0105]

[0106] Among them, T Fz For the gravity-correcting torque, α = ε + γ, θ is the wheel rotation angle around the kingpin, ε is the kingpin inclination angle, σ is the kingpin caster angle, γ is the wheel camber angle, and K is the wheel camber angle. py F is the lateral offset of the kingpin at the wheel center. z This refers to the vertical load on the tire.

[0107] The kingpin is the center of rotation of the steering wheels in a traditional car. In many cars with independent suspension, the kingpin is a virtual steering axis. During vehicle steering, the tires steer around the kingpin axis. The position of the kingpin affects the stability of the steering system and the sensitivity of the steering.

[0108] To ensure good vehicle performance and reliability, four-wheel alignment parameters need to be locked, specifically locking the wheel's kingpin rotation angle, kingpin inclination angle, caster angle, and camber angle to ensure the wheels are in the correct position. During tire steering, the lateral offset of the kingpin at the wheel center changes with the tire angle. Therefore, with the four-wheel parameters locked, the gravity-aligning moment can be calculated using the tire vertical load and tire angle. The theoretical formula for the gravity-aligning moment helps the model interpret the relationship between some vehicle parameters—namely, the tire vertical load and tire angle—and rack force as a specific function, improving the interpretability of the model during training and thus enhancing the accuracy of model predictions.

[0109] Tire drag torque is the torque generated by the resistance encountered by the tire during cornering. The vehicle must overcome the influence of tire drag torque during cornering. In one possible implementation, the theoretical formula for tire drag torque is expressed as follows:

[0110]

[0111] Among them, T Mz F is the tire drag torque, μ is the road friction coefficient, and F is the road surface friction coefficient. z ρ is the vertical load on the tire, and p is the tire pressure.

[0112] The above estimation formula is derived from tire contact patch. For the same type of vehicle or tire, the tire pressure can be set to a fixed value, such as 250 kPa. In this formula, the tire drag torque can be calculated using the road friction coefficient and the tire vertical load. The theoretical formula for tire drag torque helps the model interpret the relationship between some vehicle parameters, namely the road friction coefficient and the tire vertical load, and the rack force as a function, thereby helping the model capture the impact of changes in vehicle parameters on the rack force and further improving the accuracy of model predictions.

[0113] In this way, the computer device obtains theoretical formulas that can be used to identify the functional relationships between each set of vehicle parameters and the corresponding rack force, thereby discovering the functional relationships between some vehicle parameters and rack forces, which helps the model analyze and interpret the correlation between input and output.

[0114] After obtaining a theoretical formula that can explain the definite relationship between the model input and output, the relevant quantities of the theoretical formula are introduced into the model as intermediate variables to help the model train. In one possible implementation, the intermediate variables between each set of vehicle parameters and the corresponding rack force include: the steering lever ratio, tire resistance torque, and gravity return torque for each set of vehicle parameters.

[0115] The computer equipment obtains different values ​​for the relevant quantities based on the theoretical formulas of steering lever ratio, gravity self-centering torque and tire resistance torque according to different vehicle parameters. Before model training, it is necessary to calculate the steering lever ratio, gravity self-centering torque and tire resistance torque under different vehicle parameters.

[0116] As an example, for a rear-wheel drive vehicle, when calculating the rear wheel steering rack force, the tire vertical load is the rear axle load, the tire angle is the rear wheel angle, and the tire angular velocity is the rear wheel angular velocity. Table 2 shows the model inputs and labels for the second training sample set under this example.

[0117] Table 2

[0118]

[0119] In this way, introducing relevant quantities that can represent the functional relationship between input and output into the model as model input for training helps the model better learn the relationship between vehicle parameters and rack force, thereby improving the model's interpretability and predictive ability.

[0120] After the computer device obtains the second training sample set, it is necessary to select an appropriate supervision method for training to build a rack force prediction model. In one possible implementation, a supervised training method is used to build the rack force prediction model, including: building the rack force prediction model through the Classification and Regression Trees (CART) regression algorithm.

[0121] CART regression is a decision tree algorithm. The basic principle of building a CART regression model using computer equipment is as follows: Each input feature and its conditions are used as a decision node. A binary tree is constructed by splitting the tree layer by layer according to the decision nodes. Each leaf node of the binary tree corresponds to an output result, thus dividing the samples into M units R1, R2, ..., R3 according to the branches of the binary tree. M In this context, each unit includes input samples that satisfy the conditions corresponding to all input features under that branch, and the corresponding output result C. m When the model accuracy is 100%, C m This is the mean of the sample labels that satisfy the conditions corresponding to all input features under this branch. During the training phase, it is necessary to determine the mean of the labels based on C. m The error between the mean of the sample labels that meet the conditions under the corresponding branch and the error between the mean and the mean are used to determine the order of input features before splitting according to the decision node and the conditions of the features, so as to make the prediction result C m The mean of the sample labels that satisfy the conditions under the corresponding branch is close to the mean. The regression model f(x) can be expressed as:

[0122]

[0123] Among them, R m Let x represent the model input sample, and C represent the m-th unit. m This represents the output result corresponding to the m-th leaf node, where I represents 1 when the condition within the parentheses is met.

[0124] As an example, the specific steps for training a rack force model f(x) using a computer device are as follows:

[0125] A1: Use the squared error as C mThe error between the mean of the sample labels that meet the conditions under the corresponding branch and the error between the mean and the mean is calculated. The squared error of each unit obtained after splitting according to the decision node is then used to solve for the optimal decision node by minimizing the squared error. The optimal decision node includes the optimal splitting variable j, i.e., the optimal splitting feature in the input features, and the splitting point s, i.e., the condition corresponding to the optimal splitting feature. j and s are solved using the following formula:

[0126]

[0127] Among them, y i Let R be the mean of the sample labels in the i-th unit, R represent the unit, C represent the output result, j represent the vehicle parameters, and s represent the conditions that the vehicle parameters represented by j meet. For example, j can be the vehicle speed, and S can be whether the vehicle speed is greater than 60 km / h.

[0128] A2: Divide the cells using the selected (j,s) and determine the corresponding output values.

[0129] R1(j,s)={x|x (j) ≤s},R2(j,s)={x|x (j) >s}

[0130]

[0131] Where, N m This represents the number of all input samples in the m-th unit.

[0132] A3: Continue to repeat steps A1 and A2 for the two sub-units R1 and R2 until the stopping condition is met.

[0133] This condition can be the average of the squared errors of all units, i.e., the root mean square error, reaching a preset threshold.

[0134] A4: Divide the input samples into M units R1, R2, ..., R3 according to the optimal decision node. M In the process, a rack force prediction model f(x) is generated.

[0135] During model training, the R-squared value should be between 0.95 and 1, and the smaller the root mean square error, the better the trained model fits the actual rack force. This application provides two comparison charts of the actual and predicted rack force values ​​during the training phase, such as... Figure 3As shown, this is a comparison chart of the true and predicted values ​​provided in an embodiment of this application. The horizontal axis represents the true response of the rack force, and the vertical axis represents the predicted response of the rack force obtained based on the model. A perfect prediction value means that the predicted value of the rack force is equal to the true value of the rack force. Each observation value represents a set of vehicle parameters, that is, the true and predicted responses of the rack force under a sample. The closer the distribution of the observation values ​​is to the perfect prediction value, the better the model fit. Figure 4 As shown, this is another comparison chart of the actual value and the predicted value provided in an embodiment of this application. The computer device selected 3×10 6 In the sample, in the gray shaded area of ​​region a, a large number of true values ​​represented by black are covered by predicted values ​​represented by gray. The subtle difference between the predicted and true values ​​of the rack force can be seen from the boundary area of ​​the whole image.

[0136] The computer equipment uses the CART decision tree regression algorithm to construct the rack force prediction model. The model is represented by a binary tree structure, so that each branch of the decision tree represents a set of conditions, which makes the model construction more efficient and gives the model higher interpretability.

[0137] refer to Figure 5 The diagram illustrates a rear-wheel steering rack force prediction process according to an embodiment of this application. The computer device acquires vehicle parameters and rack forces collected from a real vehicle to form a training sample set A. The vehicle parameters are the model input, including vehicle speed, rear axle load, road friction coefficient, steering wheel angle, rear wheel angle, and rear wheel angular velocity. The model output is the rack force. Then, the computer device calculates the steering lever ratio, gravity-correcting torque, and tire drag torque under different vehicle parameters according to theoretical formulas, and adds these to the input of training sample set A to form training sample set B. The CART regression algorithm is then used for supervised training to obtain the rack force prediction model. Finally, the rack force prediction model is encapsulated into a software module and deployed in the vehicle system to predict the rack force in real time during vehicle operation.

[0138] Based on the above embodiments, this application also provides a steering rack force prediction model construction device, referring to... Figure 6 As shown, a steering rack force prediction model construction device 600 provided in this application embodiment includes an acquisition unit 601, a determination unit 602, and a construction unit 603:

[0139] The acquisition unit 601 is used to acquire a first training sample set. The model input of the first training sample set includes multiple sets of vehicle parameters. The sample label of the first training sample set includes the rack force corresponding to each set of vehicle parameters. Each set of vehicle parameters is used to identify the driving state of the vehicle. The rack force is used to control the tire steering action of the vehicle.

[0140] The acquisition unit 601 is also used to acquire a theoretical formula, which is used to identify the functional relationship between each set of vehicle parameters and the corresponding rack force;

[0141] The determining unit 602 is used to determine the intermediate variable between each set of vehicle parameters and the corresponding rack force according to each set of vehicle parameters and the theoretical formula, and add the intermediate variable as model input to the first training sample set to obtain the second training sample set. The intermediate variable is the relevant quantity of the theoretical formula.

[0142] The construction unit 603 is used to construct a rack force prediction model using supervised training based on the second training sample set. The rack force prediction model is used to predict the corresponding rack force based on vehicle parameters under different driving conditions.

[0143] In this way, by introducing intermediate variables into the training sample set through the functional relationship between some vehicle parameters and rack force, the model can explain the influence of complex changes in vehicle parameters on the corresponding rack force, and build a rack force regression prediction model with higher accuracy.

[0144] In one possible implementation, each set of vehicle parameters includes:

[0145] Vehicle speed, tire vertical load, road surface friction coefficient, steering wheel angle, tire angle, and tire angular velocity.

[0146] By using vehicle parameters that are relevant to tire steering actions as model inputs, interference from irrelevant parameters can be avoided during model training, reducing model complexity and further improving model prediction efficiency.

[0147] In one possible implementation, the theoretical formula includes:

[0148] The theoretical formulas for steering lever ratio, tire resistance torque, and gravity return torque are provided. The steering lever ratio is used to identify the relationship between tire angle and rack travel. The product of gravity return torque and tire resistance torque with rack travel is used to identify rack force.

[0149] In this way, the computer device obtains theoretical formulas that can be used to identify the functional relationships between each set of vehicle parameters and the corresponding rack force, and can discover the functional relationship between vehicle parameters and rack force, which helps the model analyze and interpret the correlation between input and output.

[0150] In one possible implementation, the theoretical formula for the steering lever ratio is expressed by the following formula:

[0151]

[0152] Where b is the steering lever ratio, δ is the tire angle, and S is the rack travel.

[0153] In one possible implementation, the theoretical formula for the gravity-correcting torque is expressed by the following formula:

[0154]

[0155] Among them, T Fz The gravity-correcting torque is given by α = ε + γ, where θ is the wheel rotation angle around the kingpin, ε is the kingpin inclination angle, σ is the kingpin caster angle, and γ is the wheel camber angle. py F is the lateral offset of the kingpin at the wheel center. z The vertical load on the tire is denoted as .

[0156] In one possible implementation, the theoretical formula for the tire drag torque is expressed by the following formula:

[0157]

[0158] Among them, T Mz The tire drag torque is μ, the road surface friction coefficient is F. z ρ is the vertical load on the tire, and p is the tire pressure.

[0159] In one possible implementation, the intermediate variables between each set of vehicle parameters and the corresponding rack force include:

[0160] The steering lever ratio, the tire drag torque, and the gravity return torque for each set of vehicle parameters.

[0161] In this way, introducing relevant quantities that can represent the deterministic functional relationship between input and output into the model as model input for training helps the model better learn the pattern between input and output, thereby improving the model's interpretability and predictive ability.

[0162] In one possible implementation, the building block is used to construct the rack force prediction model using the CART regression algorithm.

[0163] Therefore, the CART decision tree regression algorithm is used to construct the rack force prediction model, so that each branch of the decision tree represents a condition, which makes the model construction more efficient and gives the model higher interpretability.

[0164] Based on the above embodiments, this application also provides a computer device, including a processor and a memory:

[0165] The memory is used to store computer programs;

[0166] The processor is used to execute the steering rack force prediction model construction method according to the computer program.

[0167] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems or apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and relevant parts can be referred to the method section.

[0168] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for constructing a steering rack force prediction model, characterized in that, The method includes: Obtain a first training sample set. The model input of the first training sample set includes multiple sets of vehicle parameters. The sample label of the first training sample set includes the rack force corresponding to each set of vehicle parameters. Each set of vehicle parameters is used to identify the driving state of the vehicle. The rack force is used to control the tire steering action of the vehicle. Obtain the theoretical formula, which is used to identify the functional relationship between each set of vehicle parameters and the corresponding rack force; Based on each set of vehicle parameters and the theoretical formula, the intermediate variables between each set of vehicle parameters and the corresponding rack force are determined, and the intermediate variables are added as model inputs to the first training sample set to obtain the second training sample set. The intermediate variables are the relevant quantities of the theoretical formula. Based on the second training sample set, a rack force prediction model is constructed using supervised training. The rack force prediction model is used to predict the corresponding rack force based on vehicle parameters under different driving conditions.

2. The method according to claim 1, characterized in that, Each set of vehicle parameters includes: Vehicle speed, tire vertical load, road surface friction coefficient, steering wheel angle, tire angle, and tire angular velocity.

3. The method according to claim 2, characterized in that, The theoretical formulas include: The theoretical formulas for steering lever ratio, tire resistance torque, and gravity return torque are provided. The steering lever ratio is used to identify the relationship between tire angle and rack travel. The product of gravity return torque and tire resistance torque with rack travel is used to identify rack force.

4. The method according to claim 3, characterized in that, The theoretical formula for the steering lever ratio is expressed by the following formula: Where b is the steering lever ratio, δ is the tire angle, and S is the rack travel.

5. The method according to claim 3, characterized in that, The theoretical formula for the gravity-correcting torque is expressed by the following formula: Among them, T Fz The gravity-correcting torque is given by α = ε + γ, where θ is the wheel rotation angle around the kingpin, ε is the kingpin inclination angle, σ is the kingpin caster angle, and γ is the wheel camber angle. py F is the lateral offset of the kingpin at the wheel center. z The vertical load on the tire is denoted as .

6. The method according to claim 3, characterized in that, The theoretical formula for the tire resistance torque is expressed by the following formula: Among them, T Mz The tire drag torque is μ, the road surface friction coefficient is F. z ρ is the vertical load on the tire, and p is the tire pressure.

7. The method according to claim 3, characterized in that, The intermediate variables between each set of vehicle parameters and the corresponding rack force include: The steering lever ratio, the tire drag torque, and the gravity return torque for each set of vehicle parameters.

8. The method according to claim 7, characterized in that, The method of constructing a rack force prediction model using supervised training includes: The rack force prediction model was constructed using classification and regression tree CART regression algorithm.

9. A device for constructing a steering rack force prediction model, characterized in that, The device includes an acquisition unit, a determination unit, and a construction unit: The acquisition unit is used to acquire a first training sample set. The model input of the first training sample set includes multiple sets of vehicle parameters. The sample label of the first training sample set includes the rack force corresponding to each set of vehicle parameters. Each set of vehicle parameters is used to identify the driving state of the vehicle. The rack force is used to control the tire steering action of the vehicle. The acquisition unit is also used to acquire theoretical formulas, which are used to identify the functional relationships between each set of vehicle parameters and the corresponding rack force. The determining unit is used to determine the intermediate variable between each set of vehicle parameters and the corresponding rack force according to each set of vehicle parameters and the theoretical formula, and add the intermediate variable as model input to the first training sample set to obtain the second training sample set. The intermediate variable is the relevant quantity of the theoretical formula. The construction unit is used to construct a rack force prediction model using supervised training based on the second training sample set. The rack force prediction model is used to predict the corresponding rack force based on the vehicle parameters under different driving conditions.

10. A computer device, characterized in that, The computer device includes a processor and memory: The memory is used to store computer programs; The processor is configured to perform the method according to any one of claims 1-8 according to the computer program.

Images