Tire force estimation method based on physical information fusion recurrent neural network

By integrating physical information and recurrent neural networks into the vehicle dynamics model and combining vehicle state data, the problem of estimating lateral tire force in the vehicle under complex road conditions is solved, and the handling performance and safety of autonomous vehicles are improved.

CN119047303AActive Publication Date: 2024-11-29TONGJI UNIV

Patent Information

Application Number
CN202411059541.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-04
Publication Date
2024-11-29
Estimated Expiration
2044-08-04

AI Technical Summary

Technical Problem

The prior art is difficult to accurately estimate the lateral tire force of a vehicle under complex and variable road conditions, affecting the handling performance and safety of the vehicle.

Method used

Using a recurrent neural network method based on physical information fusion, the vehicle state data is collected through the Carsim simulation platform, the gated cyclic unit (GRU) network structure is designed, and the loss function is constructed in combination with the vehicle dynamic model, and the physical information and data label loss function is fused. Update network parameters to improve the accuracy of tire force estimation.

Benefits of technology

It improves the accuracy of tire force estimation and the generalization ability of the model, enhances the estimation effect under changing road conditions, and ensures the safe operation of autonomous vehicles.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119047303A_ABST
    Figure CN119047303A_ABST
Patent Text Reader

Abstract

The invention belongs to the field of tire force estimation, and particularly relates to a tire force estimation method based on a physical information fusion recurrent neural network, which comprises the following steps of: 1) designing a loss function based on a data label, and capturing an internal relation between tire force and a vehicle state sequence by utilizing a recurrent neural network algorithm; 2) combining a vehicle dynamics model, developing a loss function based on physical information, and ensuring that an estimation process follows a physical rule; and 3) fusing the two loss functions to form a tire force estimation method comprehensively considering data driving and physical constraint. The relation between tire force and the vehicle state is deeply excavated through the recurrent neural network, a loss function based on vehicle dynamics is fused, useful information is extracted from priori knowledge, and the estimation effect under variable road conditions is improved; according to the method, the flexibility of data driving and the preciseness of a physical model are combined, the accuracy of tire force estimation is improved, and the generalization ability and interpretability of the model are enhanced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of tire force estimation, and in particular relates to a tire force estimation method based on physical information fusion recurrent neural network. Background Art

[0002] In recent years, with the rapid development of social economy, the number of cars has increased day by day as a modern tool to ensure people's daily needs and improve production efficiency. The development of the automobile industry reflects the progress of my country's industrial science and technology level, and plays an indispensable and positive role in improving people's production and living quality. However, with the increase in the number of cars, traffic congestion and road traffic accidents caused by drivers' mental fatigue, drunk driving and other reasons occur from time to time, which increases the risk of road traffic system. With the rapid development of vehicle networking and artificial intelligence technology, in order to reduce traffic accidents caused by drivers, the research on autonomous driving vehicles has attracted the attention of scholars from all over the world.

[0003] The autonomous driving system is mainly composed of key modules such as the environmental perception module, planning and decision-making module, and vehicle control execution module. Autonomous driving vehicles use visual sensors and lidar sensors to collect information about the vehicle's surrounding environment and make judgments about the special driving environment around the vehicle. The sensor fusion and control module completes global and local path planning based on the information output by the environmental perception module. Finally, the control execution module solves the control strategy of the vehicle at each moment based on the output results of the path planning module, combined with the longitudinal and lateral stability and safety constraints of the vehicle, so as to achieve safe driving of the autonomous driving vehicle. Among them, vehicle motion control is at the end of the autonomous driving technology chain, carries the main behavior of the intelligent system, is a bridge for the vehicle to communicate with other modules, and plays a role in supporting and implementing the algorithm architecture of the autonomous driving system.

[0004] Accurate dynamics model is the basis for the design and optimization of intelligent driving vehicle control algorithms. In the vehicle model, the friction between the tire and the road is the main source of vehicle dynamics. In particular, the lateral tire force directly determines the vehicle's braking, steering and other handling performance. Therefore, accurate estimation of the lateral tire force is an important guarantee for the safe operation of autonomous vehicles. However, since the tire exhibits the characteristics of neither a rigid body nor an elastic element, and the friction force is related to multiple factors such as ground conditions, lateral sideslip angle, longitudinal slip, tire load, tire pressure, temperature, wear, etc., it is difficult to accurately describe this parameter through mechanism modeling. With the development of artificial intelligence, some studies have tried to use machine learning methods to establish tire force models. However, traditional tire force models based on pure data machine learning methods often do not work well in practical applications. Some recent studies are committed to enhancing the performance of machine learning models by combining physical prior knowledge, such as incorporating prior knowledge into machine learning in the form of algebraic equations or differential equations to help the model better understand the data. For the problem of tire force modeling, the modeling method of using physical information fusion neural network to improve the model's interpretability and generalization ability is of great significance to the application of the model in practical scenarios.

[0005] Aiming at the strong nonlinear characteristics of tire forces of vehicles under complex and changeable road conditions, the present invention introduces a vehicle tire force model based on a physical information fusion neural network to reduce the impact of unknown road conditions on vehicle tracking performance and safety performance, improve the estimation effect under changeable road conditions, and enhance the generalization ability and interpretability of the model. Summary of the invention

[0006] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a tire force estimation method based on physical information fusion recurrent neural network.

[0007] The technical solution of the present invention is as follows:

[0008] A tire force estimation method based on physical information fusion recurrent neural network includes the following steps:

[0009] Step 1: Collect vehicle status data and tire force data under different driving conditions based on the Carsim simulation platform;

[0010] Step 2: In the vehicle state data of step 1, select the following vehicle states related to tire forces as network input parameters:

[0011]

[0012] Where δ is the front wheel steering angle, is the yaw rate, v x ,v y are the longitudinal velocity and the lateral velocity, a x ,ay are the longitudinal and lateral accelerations, σ fl ,,σ fr ,ω rl ,σ rr They are the speeds of the left front wheel, right front wheel, left rear wheel, and right rear wheel respectively;

[0013] Step 3, perform mean-variance normalization on the vehicle status data in step 2, and divide the data into a training set and a test set;

[0014] Step 4, designing the network structure; in the present invention, the gated recurrent unit (GRU) is selected as the main recurrent neural network structure. In order to enhance the estimation effect of the model, a fully connected layer is added before the input layer as the input projection layer, and a fully connected layer is added after the output layer as the output regression layer;

[0015] Step 5: Use the network in step 4 to process the vehicle state data training set sequence to obtain the estimated value f(x i ,θ),i=T+1,…,N, T means predicting the current tire force using the state of the past T moments, and N is the number of training set data;

[0016] Step 6: Obtain the tire force estimation value f(x i ,θ), a loss function of physical information fusion is designed to update the network parameters. The design of the loss function of physical information fusion includes two parts: based on the recurrent neural network algorithm, a loss function based on data labels is designed; based on the simplified vehicle dynamics model, a loss function based on physical information is designed; the above two loss functions are superimposed to realize a tire force estimation method of physical information fusion recurrent neural network;

[0017] Step 7: Based on the loss function, update the weight parameters in the network according to the back propagation algorithm and the chain rule.

[0018] Step 8: Train the network parameters and set the maximum training round M. Repeat steps 5 to 7 within the training round until the network meets the error requirement or reaches the maximum training round, and then exit the loop.

[0019] Step 9, test the network estimation effect; obtain the tire force estimation value of the test set based on the state data of the test set and the forward propagation process of step 4. At this time, the tire force estimation value is still in the standardized scale. The tire force estimation value in the original scale is obtained through de-standardization to realize tire force estimation based on physical information fusion recurrent neural network.

[0020] Compared with the prior art, the present invention has the following beneficial effects:

[0021] In the vehicle model, the friction between the tire and the road is the main source of vehicle dynamics, especially the lateral tire force directly determines the vehicle's braking, steering and other handling performance. Therefore, accurate estimation of the lateral tire force is an important guarantee for the safe operation of autonomous vehicles. Traditional pure data-based deep learning usually requires a large amount of data to train a model with good performance and sufficient generality, and the pure data-driven model may not meet the physical laws or external constraints. Therefore, it is considered to use a neural network with physical information fusion to estimate the lateral tire force. In the network structure, in order to correctly capture the nonlinear characteristics of the tire force, in addition to considering the static mechanical characteristics of the tire-ground contact part, the dynamic characteristics caused by operations such as automatic gear shifting should also be considered. Therefore, a recurrent neural network with a simple structure and good performance, the gated recurrent unit (GRU) main network structure, is selected, and the input projection layer and output regression layer are constructed to enhance the fitting ability of the model. At the same time, an additional loss function is constructed based on the vehicle's dynamic model and fused into the recurrent neural network to improve the estimation effect under variable road conditions and enhance the generalization ability and interpretability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a flow chart of the tire force estimation method based on physical information fusion neural network proposed by the present invention;

[0023] Figure 2 is a structural diagram of a gated cycle unit according to an embodiment of the present invention;

[0024] Figure 3 This is a simplified model diagram of eight degrees of freedom of a vehicle according to an embodiment of the present invention;

[0025] Figure 4 It is a structural diagram of a tire force model based on physical information fusion recurrent neural network according to an embodiment of the present invention;

[0026] Figure 5 This is a comparison chart of the estimation results of PGRU in an embodiment of the present invention and other pure data machine learning methods;

[0027] Figure 6 This is a diagram showing the estimated effect of the algorithm of the embodiment of the present invention when the tire-road friction coefficient is 0.75;

[0028] Figure 7 This is a diagram showing the estimated effect of the algorithm of the embodiment of the present invention when the tire-road friction coefficient is 0.85;

[0029] Figure 8 This is a diagram showing the estimated effect of the algorithm of the embodiment of the present invention when the tire-road friction coefficient is 0.9. DETAILED DESCRIPTION

[0030] The technical solution provided by the present application will be further described below in conjunction with specific embodiments and accompanying drawings. The advantages and features of the present application will become more apparent with the following description.

[0031] like Figure 1 , a tire force estimation method based on physical information fusion recurrent neural network, comprising the following steps:

[0032] Step 1: Collect vehicle status data and tire force data under different driving conditions based on the Carsim simulation platform.

[0033] Step 2: In the vehicle state data of step 1, select the following vehicle states related to tire forces as network input parameters:

[0034]

[0035] Where δ is the front wheel steering angle, is the yaw rate, v x ,v y are the longitudinal velocity and the lateral velocity, a x ,a y are the longitudinal and lateral accelerations, σ fl ,,σ fr ,ω rl ,σ rr They are the speeds of the left front wheel, right front wheel, left rear wheel, and right rear wheel respectively.

[0036] Step 3: Perform mean-variance normalization on the vehicle status data in step 2 and divide it into a training set and a test set.

[0037] Step 4, design the network structure; in the present invention, the gated recurrent unit (GRU) is selected as the main recurrent neural network structure. In order to enhance the estimation effect of the model, a fully connected layer is added before the input layer as the input projection layer, and a fully connected layer is added after the output layer as the output regression layer.

[0038] The overall network structure is as follows Figure 4 As shown in the figure, the gated recurrent unit (GRU) is selected as the main recurrent neural network structure (as shown in Figure 2 In order to enhance the estimation effect of the model, a fully connected layer is added before the input layer as the input projection layer. il It can be expressed as:

[0039] x il =f i (W i x t +b i )

[0040] Among them, x t=[x t-T ,x t-T+1 ,…,x t-1 ] is the network input, where T represents the prediction of the current tire force using the state of the past T moments, and W i is the weight matrix, b i is the deviation, f i It is an activation function. In this embodiment, the activation function of the input projection layer is selected as the rectified linear unit (ReLU).

[0041] The mathematical description of the gated recurrent unit is:

[0042] z t =σ(W z x il,t +U z h t-1 +b z )

[0043] r t =α(W r x il,t +U r h t-1 +b r )

[0044]

[0045]

[0046] Among them, z t and r t By input x il,t and bias b z 、b r Through a fully connected layer, W z , W r , W h and U z , U r , U h is the weight parameter; reset gate r t Controls the hidden layer state h at the previous moment t-1 Update to the current candidate hidden layer state The degree of update gate z t Controlling the state h t-1 Update to the current hidden layer state h t The sigmoid activation function σ limits the output between 0 and 1 and is considered a logical switch. It is also calculated through a fully connected layer, using the tanh activation function to ensure that the value in the candidate hidden state remains between -1 and 1. The symbol ⊙ represents the Hadamard product, that is, element-by-element multiplication.

[0047] A fully connected layer is added to the output layer of the gated recurrent unit to enhance the model estimation effect. The mathematical representation of the output layer is:

[0048] y=W o h o,t +b o

[0049] Among them, h o,t is the hidden layer state of GRU output, W o is the weight matrix, b o is the deviation, and y is the tire force estimation data obtained after the GRU hidden layer state is regressed through the output layer.

[0050] For the training set, the forward propagation of the neural network is completed based on the above network structure. In this embodiment, T is set to 10, that is, the current tire force is predicted using the state of the past 10 moments, and the network input is x t =[x t-10 ,x t-9 ,…,x t-1 ]. t After the forward fully connected layer operation of the input projection layer, the hidden layer state x of the input projection layer is obtained il,t . Input the hidden layer state x of the projection layer il,t It is further input into the gated recurrent unit, and the long-term dependencies in the sequence are captured through the reset gate and update gate. Finally, the hidden layer state h output by the gated recurrent unit o,t After the output fully connected layer operation, h o,t The output layer contains 4 neurons, which are the estimated values ​​of the tire forces f(x t ,θ).

[0051] Step 5: Use the network in step 4 to process the vehicle state data training set sequence to obtain the estimated value f(x i ,θ),i=T+1,…,N, N is the number of training set data;

[0052] Step 6: Obtain the tire force estimation value f(x i ,θ), a loss function of physical information fusion is designed to update the network parameters. The design of the loss function of physical information fusion mainly includes two parts: based on the recurrent neural network algorithm, a loss function based on data labels is designed; based on the simplified vehicle dynamics model, a loss function based on physical information is designed; the above two loss functions are superimposed to realize a tire force estimation method of physical information fusion recurrent neural network.

[0053] The specific design and implementation method of the loss function is as follows:

[0054] (1) Vehicle dynamics model:

[0055] Consider the widely used simplified model of an eight-degree-of-freedom vehicle to establish the dynamic equations, such as Figure 3 As shown. The coordinate system takes the center of mass (G) of the vehicle as the origin, the longitudinal movement as the X-axis, and the lateral movement as the Y-axis. Based on this simplified vehicle model, the following dynamic relationship exists:

[0056]

[0057]

[0058]

[0059] Where δ is the turning angle of the front wheel, v x and v y are the longitudinal velocity and the lateral velocity, and are the longitudinal and lateral accelerations, l f and l r are the distances from the vehicle's center of mass (G) to the front and rear axles, is the yaw angle, is the yaw angular velocity, l w is the wheelbase, F xf and F xr Represents the longitudinal tire force of the front and rear wheels, F yf and F yr Represent the lateral tire force of the front and rear wheels respectively, and their mathematical description is as follows:

[0060] F xf =F xfl +F xfr , F yf =F yfl +F yfr

[0061] F xr =F xrl +F xrr , F yr =F yrl +F yrr

[0062] ΔF xf =F xfl -F xfr , ΔF yf =F yfl -F yfr

[0063] ΔF xr =F xrl -F xrr

[0064] Among them, F xj and F yj (j: = fl, fr, fl, rr) represent the longitudinal and lateral tire forces of the front left wheel, front right wheel, rear left wheel and rear right wheel respectively.

[0065] (2) Tire force estimation model based on physical information fusion recurrent neural network

[0066] The loss function of the network structure combining the vehicle physical model with the deep neural network proposed in the present invention mainly consists of two parts: a loss function based on data labels and a loss function based on physical information constraints, which are constructed as follows:

[0067] a) Loss function based on data labels

[0068] Construct a data label-based loss function based on the tire force estimation data and the reference tire force data:

[0069]

[0070] Where F = [F yfl ,F yfr ,F yrl ,F yrr ] T is the tire force data label of the training set, is the tire force estimate obtained by the network under weight θ.

[0071] b) Loss function based on physical information

[0072] In order to enable the network to obtain more information from prior physical knowledge and make the model meet basic physical constraints, the present invention integrates dynamic constraints into the loss function. Considering the complexity, we transform the dynamic model into the following form:

[0073]

[0074] According to the dynamic constraints, f PHY = 0, then the following loss function based on physical information constraints can be constructed:

[0075]

[0076] The estimated tire forces and related states should satisfy the above loss function constraints, that is, the recurrent neural network structure satisfies the dynamic physical constraints. Therefore, the loss function of the network can be written as:

[0077] loss = λ l loss label +λ ploss phy +λ r R(θ)

[0078] Among them, λ l ,λ p ,λ r is a weight scalar, where λ p It can be intuitively understood as the proportion of physical information to be added to the network. is a regularization function used to avoid overfitting problems.

[0079] Preferably, l ,λ p ,λ r Set them to 0.5, 0.5 and 0.001 respectively.

[0080] In the loss function above, the loss function based on the data label calculates the estimated value of the tire force f(x i ,θ) and the true value F i In the sum of squared errors of the training set, the expected estimated value approaches the true value; the loss function based on physical information converts the tire force estimate f(x i ,θ) into the zero equation physical constraint f of the vehicle dynamics model PHY , the expected training set f PHY It tends to 0 to meet the physical constraints and realize the training of network parameters under the joint constraints of physics and data.

[0081] Step 7: Based on the loss function, update the weight parameters in the network according to the back propagation algorithm and the chain rule.

[0082] This step is mainly implemented based on the deep learning automatic differentiation framework in PyTorch.

[0083] Step 8: Train the network parameters and set the maximum training round M. Repeat steps 5 to 7 within the training round until the network meets the error requirement or reaches the maximum training round, and then exit the loop.

[0084] Step 9, test the network estimation effect. Based on the state data of the test set and the forward propagation process of step 4, the tire force estimation value of the test set is obtained. At this time, the tire force estimation value is still in the standardized scale. After de-standardization, the tire force estimation value in the original scale is obtained to realize tire force estimation based on physical information fusion recurrent neural network.

[0085] In order to verify the superiority of the present invention, a comparative test platform is built as follows:

[0086] Test environment: Windows 10 system; Carsim simulation environment, NVIDIA GTX 1080Ti GPU; Python 3.7; PyTorch framework.

[0087] Test dataset: Vehicle driving data is collected under different driving conditions such as straight-line driving, curve driving, and double-lane lane changing.

[0088] Test index: The present invention uses MSE as a performance evaluation index. In the present invention, MSE refers to the mean square error between the tire force estimation value and the data label.

[0089] Network parameter settings: For PGRU and GRU, the input projection layer is set to a fully connected layer with 40 units, and the linear rectifier unit (ReLU) is selected as the activation function. The number of hidden layers of GRU is 20. Unlike PGRU, only the loss function based on data labels is considered in the GRU control group. The BP neural network selects a three-layer fully connected neural network with 100 hidden layer neurons; the vehicle mass is 1500 kg.

[0090] Test method: In order to verify the effect of the algorithm of the present invention, the following two groups of comparative tests are designed:

[0091] The design is compared with the pure data machine learning method. The GRU network and BP neural network without physical constraints are compared with the PGRU method proposed in this invention. The experimental results are as follows Figure 5 The yellow dotted line represents the estimation result based on the BP neural network, the red dotted line represents the estimation result of the PGRU method proposed in the present invention, and the green dotted line represents the estimation result of the GRU recurrent neural network.

[0092] The estimation effect comparison under different tire-road friction coefficients was designed. In this comparative experiment, in order to further verify the generalization ability of PGRU in estimating tire force under different road conditions, the proportion of training sets with tire-road friction coefficients of 0.75 and 0.85 was reduced to 50%, and the estimation effect under the road condition with a tire-road friction coefficient of 0.9 was tested. The division of training set and test set is shown in Table 2. The test results are shown in Figures 6 to 8 As shown in the figure, the purple dotted line represents the tire force data predicted by PGRU, and the black solid line represents the reference tire force data.

[0093] In the four sub-figures, FY_FL, FY_FR, FY_RL, and FY_RR represent the lateral tire forces of the left front wheel, right front wheel, left rear wheel, and right rear wheel, respectively.

[0094] The test results are as follows:

[0095] Table 1. Comparison of mean square error (MSE) of the estimation results of the proposed algorithm, GRU network and BP network on the test set

[0096]

[0097]

[0098] Table 2. Division of training and test sets in the estimation effect comparison experiment under different tire-road friction coefficients

[0099]

[0100] Table 3. Comparison of mean square error (MSE) of estimation results under different tire-road friction coefficients

[0101] MSE <![CDATA[F yfl ]]> <![CDATA[F yfr ]]> <![CDATA[F yrl ]]> <![CDATA[F yrr ]]> Mu=0.75 0.0113 0.0040 0.0102 0.0026 Mu=0.85 0.0027 0.0016 0.0041 0.0016 Mu=0.9 0.0029 0.0016 0.0028 0.0030

[0102] It can be seen from Table 1 that the estimation errors of PGRU in estimating the lateral force of the right front wheel and the right rear wheel are 0.0019 and 0.0018 respectively, which are significantly better than the other two groups of networks. Under the same training rounds, the estimation errors of the BP neural network on the left front wheel and the right rear wheel reached 0.01 and 0.0101 respectively, which are significantly larger than the GRU network. In addition, the overall estimation error of the BP neural network is also larger than the other two GRU-based deep neural networks. It can be inferred that considering the long-term dependency of the sequence has a better effect on the lateral tire force modeling. It can be seen from Table 3 that when the tire-road friction coefficient is 0.85, even if the proportion of the training set is reduced, PGRU can still estimate the tire force well. Outside the training domain, when the friction coefficient is 0.9, the lateral tire force estimated by PGRU also has a small estimation error. It can be shown that PGRU can better capture the characteristics of tire force and has a certain generalization ability.

[0103] It should be emphasized that the examples described in the present invention are illustrative, and their purpose is to enable people familiar with the technology to understand the content of the present invention and implement it accordingly. The present invention includes but is not limited to the examples described in the specific implementation methods. Any equivalent changes or modifications made according to the spirit of the present invention should be included in the protection scope of the present invention.

Claims

1. A tire force estimation method based on physical information fusion recurrent neural network, characterized in that: The following steps are involved: Step 1: Collect vehicle status data and tire force data under different driving conditions based on the Carsim simulation platform; Step 2: In the vehicle state data of step 1, select the following vehicle states related to tire forces as network input parameters: Where δ is the front wheel steering angle, is the yaw rate, v x ,v y are the longitudinal velocity and the lateral velocity, a x ,a y are the longitudinal acceleration and the lateral acceleration, ω fl ,,ω fr ,ω rl ,ω rr They are the speeds of the left front wheel, right front wheel, left rear wheel, and right rear wheel respectively; Step 3, perform mean-variance normalization on the vehicle status data in step 2, and divide the data into a training set and a test set; Step 4, design the network structure: select the gated recurrent unit as the main recurrent neural network structure. In order to enhance the estimation effect of the model, add a fully connected layer before the input layer as the input projection layer, and add a fully connected layer after the output layer as the output regression layer; Step 5: Execute step 4 for the vehicle state data sequence to obtain the estimated value f(x i ,θ),i=T+1,…,N, T means predicting the current tire force using the state of the past T moments, and N is the number of training set data; Step 6: Obtain the tire force estimation value f(x i ,θ), a loss function of physical information fusion is designed to update network parameters; the design of the loss function of physical information fusion mainly includes two parts: based on the recurrent neural network algorithm, a loss function based on data labels is designed; based on the simplified vehicle dynamics model, a loss function based on physical information is designed; the above two loss functions are superimposed to realize a tire force estimation method of physical information fusion recurrent neural network; Step 7: Based on the loss function, update the weight parameters in the network according to the back propagation algorithm and the chain rule. Step 8: Train the network parameters and set the maximum training round M. Repeat steps 5 to 7 within the training round until the network meets the error requirement or reaches the maximum training round, and then exit the loop. Step 9, test the network estimation effect; obtain the tire force estimation value of the test set based on the state data of the test set and the forward propagation process of step 4. At this time, the tire force estimation value is still in the standardized scale. The tire force estimation value in the original scale is obtained through de-standardization to realize tire force estimation based on physical information fusion recurrent neural network.

2. The tire force estimation method based on physical information fusion recurrent neural network according to claim 1 is characterized in that: The loss function of the network in step 7 is: loss=λ l loss label +λ p loss phy +λ r R(θ) Among them, loss labil To construct a data label-based loss function based on tire force estimation data and reference tire force data, loss phy is the loss function based on physical information constraints; l ,λ p ,λ r is the weight scalar, is a regularization function used to avoid overfitting problems.

3. The tire force estimation method based on physical information fusion recurrent neural network according to claim 2 is characterized in that: Construct a data label-based loss function based on the tire force estimation data and the reference tire force data: Where F = [F yfl ,F yfr ,F yrl ,F yrr ] T is the tire force data label of the training set, is the tire force estimate obtained by the network under weight θ.

4. The tire force estimation method based on physical information fusion recurrent neural network according to claim 2 is characterized in that: According to the simplified vehicle dynamics model, the tire force and vehicle state x are constructed i The constraint function f between PHY , and based on this constraint function, a loss function based on physical information constraints is designed:

5. The tire force estimation method based on physical information fusion recurrent neural network according to claim 2, characterized in that: λ l ,λ p ,λ r Set them to 0.5, 0.5 and 0.001 respectively.

Citation Information

Patent Citations

  • Tire force online estimation method based on neural network

    CN111708977A

  • Acceleration prediction method for longitudinal dynamics of container vehicle based on deep learning

    CN117172104A

  • Vehicle driving state estimation method

    CN117719519A

  • A method and an apparatus for estimating a friction coefficient of a vehicle wheel

    WO2023241922A1

Cited By

  • Real-time wheel center force prediction method and device and server

    CN120068282A

  • Real-time prediction method, device and server for wheel center force

    CN120068282B

  • Timing sequence neural network tire force estimation method and system under unstructured pavement

    CN120951812A

  • A timing neural network tire force estimation method and system under non-structural pavement

    CN120951812B

  • Physical information neural network rack force estimation method considering friction compensation

    CN121525538A