Tire lateral force estimation method and apparatus

Abstract

This invention provides a method and apparatus for estimating tire lateral force. The method includes: optimizing the parameters of a physical estimation model of tire lateral force and performing sensitivity analysis on key parameters of the model to obtain target stiffness parameters; determining the mapping relationship between the target stiffness parameters and the vertical load of the tire; updating the parameters of the physical estimation model of lateral force using the mapping relationship, and inputting the time-domain acceleration signal of the tire into the updated physical estimation model of lateral force to obtain a priori estimates of lateral force; extracting time-domain features from the tire's ground contact time-domain signal, fusing the time-domain features with the priori estimates of lateral force, performing nonlinear calculations on the fused features, and determining the tire lateral force estimation result based on the nonlinear calculation results and the priori estimates of lateral force. The method and apparatus of this invention, combining data-driven and model-driven approaches, improve the interpretability and accuracy of tire lateral force assessment under complex working conditions.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and apparatus for estimating tire lateral force. Background Technology

[0002] Tire lateral force is a key variable characterizing the lateral stress state of a tire, and its estimation accuracy directly affects the lateral stability control and state perception of a vehicle. However, under complex working conditions, the combined effects of tire load, lateral deviation, and road surface excitation cause the lateral force to exhibit significant nonlinear and time-varying characteristics.

[0003] In related technologies, physical models are usually used to analyze the lateral forces on tires. However, the parameters of physical models are easily affected by load changes and operating condition disturbances, making it difficult to guarantee the estimation accuracy. Related technologies also use pure data-driven methods to estimate the lateral forces on tires. This method has strong fitting ability, but poor interpretability and cross-operating condition generalization ability, resulting in inaccurate estimation results of tire lateral forces under different operating conditions. Summary of the Invention

[0004] This invention provides a method and apparatus for estimating tire lateral force, which addresses the shortcomings of existing technologies that use physical models to analyze the lateral force of tires, which are easily affected by load changes and operating condition disturbances, resulting in low accuracy of tire lateral force estimation. On the other hand, when using pure data-driven methods to estimate the lateral force of tires, the interpretability and cross-condition generalization ability are poor, resulting in inaccurate tire lateral force estimation results under different operating conditions.

[0005] This invention provides a method for estimating tire lateral force, comprising: The parameters of the physical estimation model of the lateral force of the tire are optimized to obtain the key parameters of the model, and the sensitivity analysis of the key parameters of the model is performed to obtain the target stiffness parameters; the physical estimation model of the lateral force is used to characterize the correlation between the time-domain acceleration signal of the tire and the lateral force. Determine the mapping relationship between the target stiffness parameter and the vertical load of the tire; use the mapping relationship to update the parameters of the lateral force physical estimation model, and input the time-domain acceleration signal of the tire into the updated lateral force physical estimation model to obtain the prior estimate of the lateral force; The time-domain features are extracted from the tire's ground contact time-domain signal, and the time-domain features are fused with the prior estimate of the lateral force to obtain fused features; The fusion features are subjected to nonlinear calculations, and the lateral force estimation result of the tire is determined based on the nonlinear calculation results and the prior estimate of the lateral force.

[0006] According to a tire lateral force estimation method provided by the present invention, the parameter optimization of the physical estimation model of tire lateral force to obtain key model parameters includes: An improved sparrow search algorithm was used to optimize the parameters, and the key parameters of the obtained model were obtained. The improved sparrow search algorithm is used to initialize the position of individuals in the population using the Tent chaotic mapping during the iterative optimization process, and to introduce a golden sine mechanism in the position update strategy of the population discoverer to adjust the search step size; after each generation of iterative update, the crossover and mutation mechanism of the differential evolution algorithm is used to perturb the individuals in the population.

[0007] According to the tire lateral force estimation method provided by the present invention, the step of performing sensitivity analysis on the key parameters of the model to obtain the target stiffness parameter includes: The root mean square error of the model is calculated using the single-factor analysis method when the key parameters of the model change, and the fluctuation amplitudes corresponding to the root mean square errors of the model are calculated. Different fluctuation amplitudes correspond to different sidewall lateral stiffness. The lateral stiffness of the sidewall corresponding to the maximum value among the multiple fluctuation amplitudes is determined as the target stiffness parameter.

[0008] According to the present invention, a method for estimating tire lateral force includes extracting time-domain features from the tire's ground contact time-domain signal and fusing the time-domain features with the prior estimate of the lateral force to obtain fused features, comprising: The grounding time-domain signal is input into the lateral force residual correction model; the lateral force residual correction model includes a convolutional neural network branch and a long short-term memory network branch; The grounding time-domain signal is processed using the convolutional neural network branch to obtain local transient features; the grounding time-domain signal is processed using the long short-term memory network branch to obtain time-series features. The local transient features, the temporal features, and the prior estimate of the lateral force are concatenated along the feature dimension to generate the fused features.

[0009] According to the present invention, a method for estimating tire lateral force is provided, wherein the lateral force residual correction model further includes a multilayer perceptron residual mapping layer; The step of performing nonlinear calculations on the fused features and determining the lateral force estimation result of the tire based on the nonlinear calculation results and the prior estimate of the lateral force includes: The fused features are nonlinearly mapped through the multilayer perceptron residual mapping layer to obtain the lateral force residual correction amount; Using the lateral force residual correction amount as the nonlinear calculation result, the lateral force residual correction amount is added to the lateral force prior estimate to obtain the lateral force estimation result of the tire.

[0010] According to the present invention, a method for estimating tire lateral force is provided, wherein the physical estimation model for lateral force is constructed through the following steps: Modal expansion calculations are performed on the three-dimensional physical model of the tire to obtain a coefficient matrix, which is used to characterize the equivalent circumferential distribution. The acceleration time-domain signal of the tire is converted into an equivalent distributed excitation using the coefficient matrix. The equivalent distributed excitation is then solved in conjunction with the effective rolling radius of the tire. An analytical mapping relationship between the acceleration time-domain signal and the average lateral force within a single tire cycle is constructed to obtain the physical estimation model of the lateral force.

[0011] According to a tire lateral force estimation method provided by the present invention, the method further includes: Obtain the true value of the lateral force of the tire; Calculate the root mean square error, mean absolute error, normalized root mean square error, and coefficient of determination between the estimated lateral force of the tire and the true lateral force. Based on the calculated root mean square error, mean absolute error, normalized root mean square error, and coefficient of determination, the degree of deviation between the lateral force estimation result and the true value is evaluated, and model performance evaluation results are generated.

[0012] The present invention also provides a tire lateral force estimation device, comprising: The parameter analysis module is used to optimize the parameters of the lateral force physical estimation model of the tire, obtain the key parameters of the model, and perform sensitivity analysis on the key parameters of the model to obtain the target stiffness parameters; the lateral force physical estimation model is used to characterize the correlation between the tire's acceleration time-domain signal and the lateral force. An estimation module is used to determine the mapping relationship between the target stiffness parameter and the vertical load of the tire; update the parameters of the lateral force physical estimation model using the mapping relationship, and input the time-domain acceleration signal of the tire into the updated lateral force physical estimation model to obtain the prior estimate of the lateral force. The feature fusion module is used to extract time-domain features from the tire's ground contact time-domain signal and fuse the time-domain features with the prior estimate of the lateral force to obtain fused features; The lateral force calculation module is used to perform nonlinear calculations on the fused features and determine the lateral force estimation result of the tire based on the nonlinear calculation results and the prior lateral force estimate.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the tire lateral force estimation method as described above.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the tire lateral force estimation method as described above.

[0015] The tire lateral force estimation method and apparatus provided by this invention determine reliable physical benchmark parameters through model parameter optimization, and simultaneously screen out the core stiffness parameters of the model through sensitivity analysis, providing reliable data support for subsequent reduction of computational complexity and targeted high-precision parameter correction; by establishing and utilizing a load-stiffness adaptive correction mechanism, the key parameters of the physical benchmark model are dynamically adjusted according to the working conditions; by extracting the nonlinear dynamic characteristics of tire forces under transient working conditions and introducing prior physical data, a more robust fusion feature is obtained; the actual lateral force of the tire is comprehensively evaluated by combining nonlinear calculation results and prior lateral force estimates, and by combining data-driven and model-driven approaches, the interpretability and accuracy of tire lateral force evaluation under complex working conditions are improved. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention 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 some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a schematic flowchart of the tire lateral force estimation method provided by the present invention.

[0018] Figure 2 This is a flowchart illustrating the improved sparrow search algorithm provided by the present invention.

[0019] Figure 3 This is a schematic diagram of the lateral force residual correction model provided by the present invention.

[0020] Figure 4 This is a schematic diagram of the tire lateral force estimation device provided by the present invention.

[0021] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0023] The following is combined with Figures 1-4 The present invention describes a tire lateral force estimation method and apparatus.

[0024] Figure 1 This is a flowchart illustrating the tire lateral force estimation method provided by the present invention, as shown below. Figure 1 As shown, the method includes the following steps: Step 110: Optimize the parameters of the lateral force physical estimation model of the tire to obtain the key parameters of the model, and perform sensitivity analysis on the key parameters of the model to obtain the target stiffness parameters; the lateral force physical estimation model is used to characterize the correlation between the tire's acceleration time-domain signal and the lateral force.

[0025] In this step, the lateral force physical estimation model is a mechanical analytical equation constructed based on the three-dimensional tire ring model and the time-space domain equivalence principle. It is used to establish a physical benchmark mapping between acceleration signals and lateral forces. This lateral force physical estimation model can characterize the explicit mathematical relationship between lateral acceleration and lateral forces and is used to realize the prior estimation of tire lateral forces.

[0026] In this step, parameter optimization can be achieved through relevant search algorithms or their improved versions, such as the improved sparrow search algorithm. In this embodiment, parameter optimization is performed on the physical estimation model of the lateral force of the tire to avoid the algorithm getting trapped in local optima, thereby accurately identifying the model parameters.

[0027] In this embodiment, key model parameters include free rolling radius, tire belt radius, tire belt width, tire pressure, belt linear density, sidewall radial stiffness, sidewall lateral stiffness, belt torsional stiffness, and belt out-of-plane bending stiffness.

[0028] In this step, sensitivity analysis of key model parameters can be performed using methods such as one-way factor analysis (OFAT), which can quantify the impact of a single parameter shift on the overall root mean square error (RMSE) of the model output.

[0029] In this embodiment, the target stiffness parameter can be a key factor that has the most significant impact on model error in sensitivity analysis, and can be used as a subsequent adaptive correction for working conditions.

[0030] For example, in a test scenario where a vehicle performs continuous lane-changing maneuvers on a real road, an initial lateral force physical estimator is first constructed by collecting signals using sensors inside the intelligent tires. Subsequently, the system (the main body executing the method) uses a search algorithm to find the optimal set of key model parameters in the global space. Then, the system sequentially adjusts parameters such as sidewall radial stiffness and sidewall lateral stiffness within a range of 0.5 to 1.5 times the nominal value and records the root mean square error. Analysis reveals that even a small shift in sidewall lateral stiffness can cause the error curve to exhibit a steep "V"-shaped increase. At this point, the sidewall lateral stiffness can be extracted as the aforementioned target stiffness parameter.

[0031] Step 120: Determine the mapping relationship between the target stiffness parameter and the vertical load of the tire; use the mapping relationship to update the parameters of the lateral force physical estimation model, and input the tire acceleration time-domain signal into the updated lateral force physical estimation model to obtain the prior estimate of the lateral force.

[0032] In this step, the mapping relationship can be a function relationship in which the target stiffness parameter (sidewall lateral stiffness) obtained by fitting using a linear regression method exhibits a linear decrease as the vertical load increases, which is used to characterize the nonlinear change characteristics of structural stiffness when the tire is subjected to different loads; in this embodiment, a mapping relationship between the sidewall lateral stiffness and the vertical load is established for the subsequent load correction process.

[0033] In this step, parameter updating refers to substituting the currently acquired vertical load into the mapping function to calculate a new stiffness value and replacing the nominal parameters in the physical model, so as to adapt the model to the current actual load conditions.

[0034] Specifically, in this embodiment, after substituting the above-mentioned fitted curve into the lateral force physical estimation model, the model parameters are adaptively adjusted according to the load, effectively solving the problem of accuracy imbalance of the fixed parameter model under high and low load conditions.

[0035] In this embodiment, the prior estimate of the lateral force can be obtained by inputting the real-time measured acceleration time-domain signal into the updated lateral force physical estimation model and combining it with the lateral force benchmark value calculated by Kalman filtering iteration, which is used to provide mechanically interpretable constraints and guidance for subsequent deep learning networks.

[0036] For example, in the aforementioned scenario of continuous lane-changing maneuvers, the vertical load on the tires due to vehicle tilt fluctuates drastically within a certain force range (e.g., 5500 N to 7500 N). The system pre-extracts the sidewall lateral stiffness value that minimizes model error under different loads and fits a linear attenuation mapping relationship between load and stiffness. When the vehicle sharply turns the steering wheel to the left, causing the load on the right front wheel to surge to 6500 N, the system immediately uses this mapping relationship to calculate the optimal sidewall lateral stiffness adapted to the 6500 N load and updates it to the lateral force physical estimation model. Finally, the system inputs the one-dimensional acceleration time-domain signal collected in real time by the sensors inside the tires into the model. After noise reduction through the Kalman filter prediction and update mechanism, the system calculates the prior estimate of the lateral force in the current cycle (e.g., 2800 N).

[0037] This embodiment establishes and utilizes an adaptive correction mechanism for load and stiffness to achieve dynamic adjustment of key parameters of the physical reference model according to the working conditions, effectively overcoming the accuracy imbalance problem that is prone to occur in traditional pure physical models with fixed parameters under complex load transfer conditions.

[0038] Step 130: Extract time-domain features from the tire's ground contact time-domain signal, and fuse the time-domain features with the prior estimate of the lateral force to obtain fused features.

[0039] In this step, the grounding time-domain signal refers to the local time-series data containing the complete grounding cycle, extracted using the radial acceleration peak-valley characteristics.

[0040] In this step, the corresponding temporal features can be extracted using machine learning models and deep learning models (such as convolutional neural networks and two-layer long short-term memory networks).

[0041] For example, temporal features can be local high-frequency transient features extracted by three layers of one-dimensional convolution in a convolutional neural network, or dynamic long-short-term dependent temporal features extracted by a two-layer long short-term memory network; these temporal features are used to uncover hidden dynamic patterns in signals.

[0042] In this embodiment, the result of directly splicing and combining the above-mentioned time-domain features and the prior estimates of lateral force output by the physical model along the feature dimension can be used to obtain the fused features, which can be used to achieve the complementary advantages of data-driven features and physical mechanism features.

[0043] Step 140: Perform nonlinear calculations on the fusion features, and determine the tire's lateral force estimation result based on the nonlinear calculation results and the prior estimate of the lateral force.

[0044] In this step, nonlinear computation can be achieved by performing dimensionality reduction and mapping on the fused features through a multilayer perceptron (MLP) residual mapping network consisting of multiple (e.g., 3) fully connected layers.

[0045] In this step, the nonlinear calculation result can be the lateral force residual correction amount output by the network, which is used to characterize the remaining prediction error that the physical prior model failed to fully cover when dealing with extreme and complex transient conditions.

[0046] In this embodiment, the residual correction value predicted by the network is simply added to the prior estimate of the lateral force provided by the physical model to obtain the final high-precision prediction data, namely the lateral force estimation result of the tire.

[0047] For example, in the case of continuous lane changing, the system inputs the fused features into the three-layer fully connected MLP at the end of the residual correction network for nonlinear calculation. At this time, the network does not directly fit the true absolute value of the tire's lateral force, but evaluates the calculation blind spot of the physical model and outputs the "lateral force residual correction amount" that the physical model has under-calculated or over-calculated (for example, it calculates that there is a nonlinear error compensation of +120 Newtons due to the rapid change in sideslip). Finally, the system adds this 120 Newton residual correction amount to the 2800 Newtons prior estimate calculated by the physical model to obtain the final tire lateral force estimate of 2920 Newtons at the current moment.

[0048] The tire lateral force estimation method provided in this invention determines reliable physical benchmark parameters through model parameter optimization and selects core stiffness parameters of the model through sensitivity analysis, providing reliable data support for subsequent reduction of computational complexity and targeted high-precision parameter correction. By establishing and utilizing a load-stiffness adaptive correction mechanism, the key parameters of the physical benchmark model are dynamically adjusted according to the working conditions. By extracting the nonlinear dynamic characteristics of tire forces under transient working conditions and introducing prior physical data, a more robust fusion feature is obtained. The actual lateral force of the tire is comprehensively evaluated by combining the nonlinear calculation results and the prior lateral force estimates. By combining data-driven and model-driven approaches, the interpretability and accuracy of tire lateral force evaluation under complex working conditions are improved.

[0049] In some embodiments, parameter optimization is performed on the physical estimation model of the lateral force of the tire to obtain key model parameters, including using an improved sparrow search algorithm to optimize parameters and obtain key model parameters; wherein, the improved sparrow search algorithm is used to initialize the position of individuals in the population using the Tent chaotic mapping during the iterative optimization process, and introduces a golden sine mechanism in the position update strategy of the population finder to adjust the search step size; after each generation of iterative update, the crossover and mutation mechanism of the differential evolution algorithm is used to perturb the individuals in the population.

[0050] It should be noted that the Sparrow Search Algorithm (SSA) is a swarm intelligence optimization method. This algorithm searches for the optimal solution by simulating the foraging and vigilance behaviors of sparrows. The algorithm has a simple structure, few parameters, and good global search capability and convergence performance. Specifically, SSA divides the population into three categories: discoverers, followers, and vigilants. Discoverers are responsible for searching potentially superior areas, followers update their positions based on population information, and vigilants sense risks and guide the population to adjust its search direction. Through the collaborative effect between these different roles, the algorithm achieves a good balance between global exploration and local exploitation.

[0051] Specifically, assuming by A population consisting of only sparrows, in Position matrix in 3D search space X Defined as: ; in, Indicates the first Only sparrows in the first The position coordinates in 3D space; a vector consisting of the fitness values ​​of all sparrows in the population. Represented as: ; in, The set of fitness scores for all individuals. This is the fitness function, used to evaluate the quality of each sparrow's position.

[0052] The discoverer location update strategy provided in this embodiment includes: the discoverer, as an individual with high energy reserves in the population, undertakes the task of exploring food sources, and its location update formula is: ; in, This represents the current iteration number; , ; It is a random number; This represents the maximum number of iterations. For each element that is 1 3D row vector; Random numbers that follow a standard normal distribution; This is a warning value, and its range is [value range missing]. ; As a safety threshold, it is usually set to .

[0053] In this embodiment, when When this indicates the foraging environment is safe, the finder can search the current area extensively; when This indicates a predator threat, and the population needs to quickly move to other safe areas to forage.

[0054] The follower location update strategy provided in this embodiment includes: the follower adjusts its foraging strategy by monitoring the location changes of the discoverer, and the update formula is as follows: ; in, For the first The optimal position occupied by the producer in the next iteration; This represents the current worst-case position globally; matrix for A 3D matrix whose elements randomly take values ​​of 1 or -1. ;when When the follower has not received food, it indicates that the follower needs to move to another area; when At this time, the follower performs a local search in the vicinity of the current optimal region.

[0055] The vigilant location update strategy provided in this embodiment includes: vigilant populations typically comprise 10% of the population. 20%, its position update formula is: ; in, The step size control parameter follows a normal distribution with a mean of 0 and a variance of 1. It is a random number; To avoid extremely small constants with a denominator of zero; This represents the current fitness value of the sparrow. and These are the current worst and best fitness values, respectively; when This indicates that the vigilant individual is on the edge of the population and has detected danger, requiring it to move to an optimal location to reduce the risk of predation.

[0056] Based on this, this embodiment constructs an improved sparrow search algorithm through the following steps: (1) Population initialization based on Tent chaotic mapping; In standard SSA, the population is typically initialized randomly. While this is convenient, randomly generated individuals are prone to uneven distribution, affecting population diversity and hindering the global search in the early stages of the algorithm. To address this issue, this embodiment uses Tent chaotic mapping to generate the initial population.

[0057] In this embodiment, since chaotic sequences have good ergodicity and uniformity, the Tent map can improve the distribution quality of individuals in the solution space, thereby providing more reasonable initial conditions for subsequent optimization; the expression for the Tent chaotic map is as follows: ; in: For control parameters, ; This represents the number of iterations. It is a chaotic variable.

[0058] In this embodiment, based on the distribution map and histogram of the generated sequences from the Tent chaotic mapping plotted in the preliminary experiment, it can be seen that the chaotic sequences in It exhibits a relatively uniform distribution within the interval, has good ergodicity, and can more fully cover the solution space.

[0059] (2) Discoverer position update based on the golden sine strategy; In the traditional SSA search process, while discoverers converge quickly, they are prone to prematurely concentrating in local regions in the early stages of iteration, thus affecting the accuracy of subsequent optimization. To enhance the global exploration capability during the search process while also considering local development effects, this embodiment introduces a golden sine strategy to improve the discoverer's position update method. This strategy combines the golden section concept with the characteristics of sinusoidal perturbation, which can dynamically adjust the search direction and step size during iteration, enabling the population to improve its ability to explore potentially superior regions while maintaining a certain search range.

[0060] Specifically, the improved discoverer location update formula is as follows: ; in, This represents the number of iterations. This indicates the current location of the sparrow. , It is a random number. , ; For the first The globally optimal position obtained by substitution; This is a warning value. ; For safety values, ; These are random numbers that follow a standard normal distribution. It is a matrix whose elements are all 1s.

[0061] coefficient and The calculation formula is: ; in, , The coefficient of the golden sine division; It is the golden ratio. ; , The initial value is set to and , , It dynamically adjusts as the target value changes, thereby driving , Dynamically updated.

[0062] The improved discoverer location update process can maintain a larger search step size in the early stage of iteration to enhance global exploration capabilities, and gradually narrow the search range in the later stage of iteration to strengthen local development capabilities.

[0063] (3) Mutation strategies based on differential evolution; In the later stages of SSA iteration, as the population gradually converges and individual differences decrease, search stagnation is likely to occur, thereby weakening the algorithm's ability to escape local optima. Differential Evolution (DE) continuously introduces differential perturbations during the evolutionary process through mutation and crossover operations, which can effectively maintain population diversity. Based on this, this embodiment introduces the mutation and crossover mechanism of differential evolution into the SSA iteration process to enhance the algorithm's global search capability in the later stages.

[0064] Specifically, in each iteration, after the individuals in the population have completed the exploration and development phase of the original SSA, a DE operation is introduced for each individual. The specific process is as follows: 1. Mutation operation; Randomly select three different individual indexes And generate a mutation vector according to the difference mutation formula, the corresponding mutation formula is: ; in: For the first The individual in the first The mutation vector of the generation; , , Three different individuals were randomly selected; The scaling factor is set to 0.5; the above mutation methods can introduce new genetic information into the population, break the limitations of the original individuals, and enhance the population's exploration capabilities.

[0065] 2. Cross-operation; After the mutation operation is completed, this embodiment performs a crossover operation between the mutated vector and the original individual. First, a dimension index is randomly selected. And based on the crossover probability Generate cross individuals , its first The generation rules for dimensional components are as follows: ; in, The crossover probability has a range of values. ; for Uniformly distributed random numbers within an interval; For the new individual after crossover Dimensional components; For the original individual's first Dimensional components.

[0066] The above crossover strategy retains some structural features of the original individuals while introducing information from the mutation vector, thus achieving effective recombination of the solution vector.

[0067] 3. Select the operation; A selection operation is performed after the crossover operation. The individuals after the crossover are compared. and primitive individuals Fitness values ​​are used to retain better individuals using a greedy selection strategy: If the individuals after crossover Better adaptability, that is Then update the position of the individual in the population to Otherwise, retain the original individual.

[0068] Figure 2 This is a flowchart illustrating the improved sparrow search algorithm provided by the present invention. Figure 2 In the illustrated embodiment, the method includes the following steps: Step 1: Population initialization; The sparrow population is initialized using the Tent chaotic mapping, and the objective function and population size are set. Problem Dimension Maximum number of iterations Upper and lower bounds of the search range ( , ) and other parameters are set to generate the initial population location.

[0069] Step 2: Calculate and sort the fitness values; Calculate and sort the fitness values ​​of all individuals, find the positions of the best and worst individuals, and record the fitness value sequence.

[0070] Step 3: Determine the iteration conditions; Determine if the current iteration number T is less than or equal to the maximum iteration number Tmax. If not, output the optimal solution and terminate; if yes, continue with the subsequent steps.

[0071] Step 4: Update the discoverer's location; The golden sine strategy is introduced to improve the position update formula of the discoverer. The position of the discoverer is updated according to formulas (16), (17), and (18), which enhances the global search capability of the algorithm.

[0072] Step 5: Update follower positions; The position of the follower is updated according to formula (17) to make it move closer to the optimal position.

[0073] Step 6: Update the location of the vigilant; Some sparrows have an alertness ability and update their position through formula (18), thereby enhancing the local development ability of the population.

[0074] Step 7: Execute the difference mutation and greedy strategy; Perform differential mutation operation on the updated sparrow positions and adopt a greedy strategy to retain the better solution to further improve the optimization performance of the algorithm.

[0075] Step 8: Return to step 2, recalculate the fitness value and proceed to the next iteration until the termination condition is met.

[0076] In this embodiment, after introducing Tent chaotic mapping initialization, the golden sine strategy, and the difference mutation operator, ISSA exhibits better optimization accuracy and more stable convergence performance on most test functions. In particular, for unimodal functions and most multimodal functions, ISSA's optimal value, average value, and standard deviation are all superior to traditional search algorithms.

[0077] The tire lateral force estimation method provided in this invention solves the defects of traditional physical model parameter identification process, which is prone to getting trapped in local optima and has insufficient convergence accuracy, by deeply integrating Tent chaotic initialization, golden sine step size adjustment and differential evolution mutation operation, and provides reliable underlying model key parameters for subsequent calculations.

[0078] In some embodiments, sensitivity analysis is performed on key model parameters to obtain target stiffness parameters, including: using single-factor analysis to calculate multiple root mean square errors of the model as the values ​​of key model parameters change, and calculating multiple fluctuation amplitudes corresponding to the multiple root mean square errors of the model, with different fluctuation amplitudes corresponding to different sidewall lateral stiffness; determining the sidewall lateral stiffness corresponding to the maximum value among the multiple fluctuation amplitudes as the target stiffness parameter.

[0079] In this embodiment, one-factor analysis (OFAT) refers to a testing strategy that adjusts only the value of a certain parameter while keeping the other parameters unchanged at their nominal values ​​when analyzing a certain parameter. It is used to isolate and quantify the independent influence of a single variable on the accuracy of a physical benchmark model.

[0080] In this embodiment, the root mean square error (RMSE) of the model is a quantitative indicator of the absolute deviation between the lateral force estimated by the physical model and the actual measured value, and is used to evaluate the model accuracy under specific parameter values.

[0081] In this embodiment, the fluctuation amplitude refers to the size of the range of change of the root mean square error as the parameter value is scaled, which is used to characterize the sensitivity of the model to a specific parameter; the target stiffness parameter refers to the stiffness variable (i.e., the sidewall lateral stiffness) that causes the largest error fluctuation amplitude among all the key parameters tested, which is used as the core parameter for subsequent dimensionality reduction optimization and load adaptive correction.

[0082] In this embodiment, the accuracy of the physical model is highly dependent on the selection of model parameters. Under actual working conditions, the tire's ground contact characteristics and structural stiffness will change nonlinearly with the change of vertical load. In order to construct a high-precision physical benchmark model, this embodiment first uses the One-Factor-At-A-Time (OFAT) method to perform sensitivity analysis on the key parameters of the model.

[0083] Specifically, select the belt linear density Radial stiffness of sidewalls Lateral stiffness of sidewalls belt torsional stiffness Out-of-plane bending stiffness of the belt and the polar inertia of the belt As an input factor, when analyzing the sensitivity of a certain parameter, only that parameter is adjusted, setting its value to 0.5 to 1.5 times the nominal value, while keeping the other parameters unchanged, and then substituting that parameter into the model to calculate the corresponding model error.

[0084] In this embodiment, the root mean square error (RMSE) is used to characterize the model error. The greater the fluctuation of RMSE with parameter changes, the more sensitive the parameter is to the model output.

[0085] In this embodiment, sensitivity curves for each parameter under different load conditions were plotted through preliminary experiments, and analysis of the sensitivity curves revealed that the lateral stiffness of the sidewall... The curve exhibits an extremely steep "V" shape, indicating that even a small shift in this parameter can lead to a dramatic increase in model error, making it a core factor affecting model accuracy. (Out-of-plane bending stiffness of the belt) The curve also shows a certain slope, indicating that it has a certain modulation effect on the model output. It can be seen that the lateral stiffness of the sidewall always shows the strongest sensitivity, which fully demonstrates that it is the most critical parameter that determines the performance of the physical model. Therefore, in order to balance the model calculation efficiency and correction accuracy, this embodiment only establishes the mapping relationship between the lateral stiffness of the sidewall and the vertical load in the subsequent load correction process.

[0086] Specifically, the optimal sidewall lateral stiffness that minimizes model error under different loads is first extracted. The value increases with the load. It exhibits a clear linear decay trend. In this embodiment, a linear regression method is used for fitting, and the resulting correction function is expressed as follows: ; Among them, the coefficient of determination of the above-mentioned fitted curve The value reached 0.946, which indicates a strong correlation between the stiffness parameter and the working conditions.

[0087] The tire lateral force estimation method provided in this invention calculates the root mean square error of multiple models when the key parameters of the model change using a single-factor analysis method, and calculates multiple fluctuation amplitudes corresponding to the multiple root mean square errors of the models. Finally, the sidewall lateral stiffness corresponding to the maximum value among the multiple fluctuation amplitudes is taken as the target stiffness parameter. This method eliminates the interference of low-sensitivity parameters, accurately selects the core parameters that play a decisive role in the accuracy of the model, and reduces the computational complexity of subsequent adaptive updates of the physical model.

[0088] In some embodiments, time-domain features are extracted from the tire's ground contact time-domain signal, and the time-domain features are fused with the prior estimate of lateral force to obtain fused features. This includes: inputting the ground contact time-domain signal into a lateral force residual correction model; the lateral force residual correction model includes a convolutional neural network branch and a long short-term memory network branch; processing the ground contact time-domain signal using the convolutional neural network branch to obtain local transient features; processing the ground contact time-domain signal using the long short-term memory network branch to obtain temporal features; and concatenating the local transient features, temporal features, and prior estimate of lateral force along the feature dimension to generate fused features.

[0089] In this embodiment, the lateral force residual correction model refers to a hybrid network architecture that combines deep learning and physical constraints. This model is used to correct the nonlinear errors generated by traditional physical models.

[0090] In this embodiment, the Convolutional Neural Network (CNN) branch can perform one-dimensional convolution and max pooling operations to capture high-frequency, bursty local transient features from a one-dimensional ground signal.

[0091] In this embodiment, the Long Short-Term Memory (LSTM) network branch is a time-series processing channel composed of stacked LSTM networks, used to model the evolution of signals and obtain the sequential characteristics of tire force in the time dimension.

[0092] In this embodiment, the operation of directly concatenating the feature vectors output by different network branches with the scalar values ​​obtained from the physical model in the channel dimension yields the corresponding fused features, which contain deep data information and physical and mechanical information.

[0093] It should be noted that in actual driving scenarios, especially when vehicles are changing lanes, the tire lateral force exhibits typical transient nonlinear variation characteristics. Its time-series curve approximates a sinusoidal fluctuation and has obvious dynamic evolution patterns. Although relying solely on physical models can provide a mechanically meaningful prediction benchmark, it is difficult to accurately capture the transient dynamic response of the lateral force. Therefore, this embodiment designs a residual correction architecture that integrates physical information and data-driven approaches, namely the lateral force residual correction model, to accurately estimate the tire lateral force and fully leverage the complementary advantages of physical models and deep learning.

[0094] In this embodiment, the lateral force residual correction model adopts an overall architecture of "parallel feature extraction - fusion mapping - residual superposition". The network takes the sensor grounding time-domain signal as input and performs parallel feature extraction through CNN and LSTM branches respectively: the CNN branch extracts the local transient features of the grounding signal through three layers of one-dimensional convolution and pooling operations; the LSTM branch performs time-series modeling on the same time-domain signal to capture the long-term and short-term dependencies of the dynamic changes of lateral force; at the same time, the lateral force prior estimate output by the physical model is used as the third input to directly participate in feature concatenation, providing physical prior constraints for the network.

[0095] Figure 3 This is a schematic diagram of the lateral force residual correction model provided by the present invention. Figure 3 In the embodiment shown, the lateral force residual correction model is constructed through the following steps: Step 1, Input Layer: The network input is the one-dimensional time-domain signal of tire ground contact acquired by the sensor. Let the length of the input sequence be... Then the input sequence can be represented as: ; The original input signal is preprocessed by normalization, and the processed sequence is simultaneously input into the CNN branch and the LSTM branch for parallel feature extraction.

[0096] Step 2, CNN Feature Extraction Layer: The CNN module consists of three one-dimensional convolutional layers, which extract local transient features from the input grounded time-domain signal; The output of a convolutional layer can be represented as: ; ; in, Indicates the first Convolutional layer output, initial input ; Indicates the first Layer convolution kernel weights; Indicates the bias term; This represents a one-dimensional convolution operation; Each convolutional layer is followed by max pooling to reduce feature dimensionality. The pooling process is represented as follows: ; Third-level pooling output After mapping through fully connected layers, the CNN feature vectors are obtained. That is, local transient characteristics: .

[0097] Step 3, LSTM timing modeling layer: The LSTM module consists of two stacked LSTM networks. The LSTM takes the input time domain signal as input. Using this as input, the temporal pattern of the dynamic changes in tire lateral force is modeled to capture the long-term and short-term dependencies of the grounding signal in the time dimension; specifically, The output of the constantly hidden state is: ; in, express Always hide your status; This indicates the hidden state in the previous moment; express Time-based input; In this embodiment, the second layer of the dual-layer LSTM takes the hidden state sequence of the first layer as input and takes the hidden state at the last time step. The LSTM feature vectors are obtained after mapping through a fully connected layer. In other words, time sequence features: .

[0098] Step 4, Feature Concatenation Layer: Concatenates the CNN feature vectors. LSTM feature vectors and the prior estimate of the lateral force output by the physical model. By concatenating along the feature dimensions, the fused features are obtained. : ; in, This indicates a concatenation operation along the feature dimension. The lateral force estimate output by the physical model at the current moment is directly introduced into the feature splicing layer to provide physical prior constraints for the residual correction network.

[0099] The tire lateral force estimation method provided in this embodiment of the invention constructs a fusion architecture that combines parallel CNN and LSTM dual feature extraction branches with direct physical numerical data. This ensures the model's sensitivity to high-frequency transient changes in tire response and enables it to learn the temporal evolution of mechanical states. It achieves deep complementarity between physical prior knowledge and data-driven features, thereby improving the error prediction accuracy of the residual network.

[0100] In some embodiments, the lateral force residual correction model further includes a multilayer perceptron residual mapping layer; performing nonlinear calculations on the fused features and determining the lateral force estimation result of the tire based on the nonlinear calculation results and the prior lateral force estimate, including: performing nonlinear mapping on the fused features through the multilayer perceptron residual mapping layer to obtain the lateral force residual correction amount; using the lateral force residual correction amount as the nonlinear calculation result, adding the lateral force residual correction amount and the prior lateral force estimate to obtain the lateral force estimation result of the tire.

[0101] In this embodiment, the multilayer perceptron residual mapping layer (MLP residual mapping layer) consists of multiple fully connected hidden layers, which are used to perform deep nonlinear transformation and feature parsing on high-dimensional fused features containing both physical and data information.

[0102] In this embodiment, the lateral force residual correction refers to the specific deviation compensation value finally output by the network after nonlinear mapping. It is used to accurately quantify and characterize the residual error (i.e. nonlinear unmodeled dynamics) that the pure physical model fails to accurately cover when facing complex time-varying road surface excitation and severe load transfer.

[0103] In this embodiment, the three pieces of information—lateral force residual correction, temporal features, and prior lateral force estimates—are fused through feature concatenation and then input into a three-layer fully connected residual MLP for nonlinear mapping. The residual correction is output and finally superimposed with the physical model prediction to obtain an accurate tire lateral force estimation result.

[0104] Specifically, let's continue with the construction steps of the aforementioned lateral force residual correction model: Step 5, Residual MLP Mapping Layer: The residual MLP consists of three fully connected hidden layers, which fuse feature vectors. Perform nonlinear mapping and output residual correction; The output of the hidden layer is: ; in, ; For the first Layer weight matrix; It is the bias vector; The residual correction amount is obtained by linear mapping of the output layer. It can be expressed by the following formula: ; in, This represents the output layer weight matrix; This represents the output layer bias vector.

[0105] Step 6, Output Layer: Correct the output of the residual MLP. Compared with the physical model predictions (i.e., the prior estimates of lateral forces). By superimposing the results, we obtain the estimated lateral force of the tire: ; in, This represents the final estimated tire lateral force. This represents the prior estimate of the lateral force; This indicates the result of the lateral force estimation.

[0106] The tire lateral force estimation method provided in this embodiment of the invention designs a lateral force residual correction model. The model utilizes the residual superposition mechanism and only needs to learn the distribution law of the prediction error of the physical model, rather than directly fitting the absolute value of the lateral force. This effectively reduces the learning difficulty of the network and improves the prediction accuracy and generalization ability of the model under complex working conditions.

[0107] In some embodiments, the lateral force physical estimation model is constructed through the following steps: modal expansion calculation is performed on the three-dimensional physical model of the tire to obtain a coefficient matrix, which is used to characterize the equivalent circumferential distribution; the acceleration time-domain signal of the tire is converted into an equivalent distributed excitation using the coefficient matrix, and the equivalent distributed excitation is solved in combination with the effective rolling radius of the tire, and an analytical mapping relationship between the acceleration time-domain signal and the average lateral force in a single tire cycle is constructed to obtain the lateral force physical estimation model.

[0108] In this embodiment, the tire lateral force can be estimated in real time by constructing an explicit mathematical relationship between lateral acceleration and lateral force; specifically, based on the tire three-dimensional ring model theory, the lateral acceleration measured on the inner liner of the smart tire can be expressed as: ; in, The angular velocity of the tire. Indicates the lateral displacement of the belt layer nodes. The second derivative with respect to the corner position of the node.

[0109] In this embodiment, the tire control equations are solved using modal expansion techniques, which can... Position of the upper corner of the belt layer circumference at any moment The lateral displacement at a point is expressed in series form, and is represented by the following formula: ; in, These are coefficients related to model parameters and modal order; To act on the circumference of the belt layer Lateral distribution excitation at the location; This represents the number of nodes within the grounding area; The modal order is denoted by .

[0110] In this embodiment, taking the second derivative of the above equation with respect to angular position yields the following equation: ; If using angle step size Discretize the belt layer circumference as If there are nodes, then at any time... Below, all nodes on the circumference Can constitute 3D column vector All within the grounding area constitute 3D column vector The two satisfy the following linear relationship: ; in, for The coefficient matrix, whose elements are determined by the model parameters and modal coefficients, can be calculated using the following formula: ; On the other hand, tire lateral force It can be expressed as the integral form of the distributed excitation within the grounding region, as follows: ; in, The effective rolling radius of the tire. For all elements are 1 3D row vectors.

[0111] It should be noted that if the distribution is uniform along the tire circumference... While a lateral acceleration sensor is available, its cost and implementation are significant. To address this issue, under stable operating conditions, the time-series data measured during one revolution of a single belt node can be equivalent to the spatial distribution data of all nodes on the entire circumference at a given instant. Based on this equivalence, this embodiment utilizes the sampling period of the smart tire signal. and tire speed The estimated number of data points within one tire cycle is expressed as: ; in, The floor sign is used for rounding down; correspondingly, the angle step size on the equivalent circle is represented as... .

[0112] In this embodiment, a grounding area identification algorithm can be used to utilize the peak and valley characteristics of the radial acceleration signal to help extract effective information about the lateral acceleration.

[0113] Specifically, by identifying the left and right peak points of the radial acceleration signal, the number of data points corresponding to the time span between the two peak points is used as the equivalent number of grounding nodes. Compared to directly calculating the interval of peak points with the same sign from the lateral acceleration signal, this method utilizes the high signal-to-noise ratio of the radial acceleration signal, improving the robustness of grounding area identification. Simultaneously, it extends forward and backward from the valley point (grounding center) of the radial acceleration signal as a reference. A total of [number] data points were extracted, encompassing the entire tire cycle. This extraction method, which uses lateral acceleration data points, ensures that the selected signal segment fully covers the tire's ground contact process.

[0114] In this embodiment, the correspondence between the equivalent circumferential parameters and the actual measured signal is shown in Table 1: Table 1. Comparison of Equivalent Parameters in the Time-Space Domain Regarding Table 1 above, define 3D column vector and 3D column vector Let represent the second derivative of the lateral displacement on the equivalent circle and the distributed excitation, respectively; by analogy with the spatial domain model, an equivalent relationship in the time domain can be established, as shown in the following equation: ; Among them, matrix The equivalent distributed excitation can be obtained by substituting the equivalent parameters into the model equations. , is represented as: ; Therefore, the average lateral force on the equivalent circle can be expressed as: ; In the formula, for A unit row vector. Combining the above two equations, we obtain the explicit expression for the lateral force estimator: This model establishes a linear mapping relationship between lateral acceleration signals and lateral forces, vector... It can be obtained directly from the measured signal after filtering. It should be noted that when using a single sensor, the estimated value can only be updated once per tire cycle. It reflects the average lateral force over one cycle.

[0115] To fully utilize the measurement information from the accelerometer and effectively suppress the interference of measurement noise on the estimation results, a Kalman filter algorithm is introduced into the constructed lateral force estimator. As a classic recursive optimal estimation method, the Kalman filter can fuse the system dynamic model and real-time measurement data within a state-space framework, achieving optimal estimation of state variables by minimizing the mean square error of the estimation error. This method has been widely verified and applied in fields such as vehicle dynamics state observation and integrated navigation systems. Since the aforementioned estimator model maintains the linear relationship between state variables and observations, the standard linear Kalman filter algorithm can be directly used, avoiding the computational complexity introduced by nonlinear filtering methods.

[0116] The filtering process includes two key steps: prediction and updating. The optimal estimate of the lateral force is achieved through iterative calculation.

[0117] The first step is the prediction step, which involves making a prior prediction for the current time step based on the posterior estimate from the previous time step: in, express Prior estimates of the lateral force at time step; The prior estimation error covariance reflects the degree of uncertainty in the prediction result; The process noise covariance describes the inherent error of the system model.

[0118] In this embodiment, after obtaining new measurement data, the Kalman gain is calculated and the prior estimate is corrected: ; in, For Kalman gain, This is the posterior estimate obtained after fusing measurement information, which is the optimal estimate of the lateral force at the current moment; For the posterior error covariance; To measure the noise covariance, the reliability of sensor measurements is quantified.

[0119] In this embodiment, the filtering algorithm uses error covariance. and , The interaction between these factors achieves a dynamic balance between predicted and measured values. When measurement noise is high, the gain... Decreasing the filter makes it more reliant on the predictions; conversely, increasing its size makes it more dependent on the measurement data. This adaptive adjustment mechanism enables the estimator to maintain good robustness and accuracy under different operating conditions.

[0120] The tire lateral force estimation method provided in this invention uses differential equation modal expansion and spatiotemporal equivalence transformation mechanism to map local acceleration signals into overall tire lateral force. This physical modeling process solves the problem that a single sensor cannot simultaneously acquire full-circumference spatial distribution data, and provides a physical benchmark model with high mechanical interpretability and computational efficiency for subsequent introduction of deep networks for residual compensation.

[0121] In some embodiments, the tire lateral force estimation method further includes: obtaining the true value of the tire's lateral force; calculating the root mean square error, mean absolute error, normalized root mean square error, and coefficient of determination between the estimated lateral force and the true value of the tire; and evaluating the degree of deviation between the estimated lateral force and the true value based on the calculated root mean square error, mean absolute error, normalized root mean square error, and coefficient of determination, thereby generating a model performance evaluation result.

[0122] This embodiment uses experimentally collected data to conduct two types of tests: bench tests and real vehicle tests. The bench tests use transient condition data (SS) and steady-state condition data (TS), and extract the model input and corresponding lateral force labels from these datasets as outputs for model training and validation. The real vehicle tests use real vehicle condition datasets (RV-1, RV-2) as external test sets to directly validate the model without retraining it, and to evaluate the generalization ability and robustness of the lateral force physical estimation model under real road conditions.

[0123] In this embodiment, the raw signals collected in bench tests and real vehicle tests are inevitably affected by various factors such as sensor noise, electromagnetic interference, and road surface excitation, resulting in a large number of high-frequency noise components in the data. In order to effectively extract the real tire dynamics characteristics, this embodiment uses a low-pass filter to denoise the raw data.

[0124] It should be noted that the low-pass filter (the cutoff frequency can be 250Hz) is selected based on the fact that the main dynamic response frequency components of tire lateral force are usually concentrated within 200Hz. Therefore, setting the cutoff frequency to 250Hz can effectively filter out high-frequency interference noise and retain the dynamic characteristics of the effective signal to the greatest extent, thus avoiding signal distortion.

[0125] In this embodiment, due to the different physical meanings of the input features, the numerical ranges and distribution characteristics of the variables vary significantly. If the original data is used directly to train the model, variables with larger values ​​will dominate during gradient descent, leading to difficulties in model convergence and decreased generalization performance. Therefore, this embodiment requires standardization of each feature variable to eliminate the influence of units and make them comparable. The standardization calculation method is as follows: ; in, This represents the standardized data. This represents the original data after filtering. The mean of this variable. Let be the standard deviation of the variable. After the above transformation, each characteristic variable is transformed into a standard normal distribution with a mean of 0 and a standard deviation of 1.

[0126] In addition, to avoid leakage of test set information, the standardized parameters of all datasets ( and All data are obtained based on training set statistics to ensure strict consistency in data processing. The standardized data distribution in this embodiment is more balanced, which helps to accelerate the convergence of the neural network and improve the estimation accuracy of the model.

[0127] To comprehensively evaluate the model's prediction accuracy and fitting performance, this embodiment employs multiple evaluation metrics, including root mean square error (RMSE), mean absolute error (MAE), normalized root mean square error (NRMSE), and coefficient of determination (COP). These indicators reflect the degree of deviation between predicted and actual values ​​from different perspectives, providing a quantitative basis for the comprehensive evaluation of model performance; the calculation methods for each indicator are as follows:

[0128] (1) The root mean square error (RMSE) is expressed as: ; (2) The Mean Absolute Error (MAE) is expressed as: (3) The normalized root mean square error (NRMSE) is expressed as: ; (4) Coefficient of Determination ) is represented as: ; in, The total number of samples, Indicates the first The true value of each sample Indicates the first The predicted value for each sample, The average of all true values. It represents the set of all real values.

[0129] This embodiment describes the predictive performance of the model from different dimensions through the above evaluation indicators. Among them, RMSE, MAE and NRMSE mainly reflect the absolute size of the prediction error. The smaller the value of these three indicators, the higher the accuracy of the prediction model. Used to evaluate the predictive performance of the model, its value ranges from [0, 1]. The closer the value is to 1, the better the model fits the data and the stronger its predictive ability. By comprehensively analyzing these evaluation indicators, we can make a comprehensive and objective judgment on the predictive performance and practical application value of the model.

[0130] The tire lateral force estimation method provided in this invention establishes an evaluation mechanism that includes absolute error verification, relative error comparison, and global fit testing, thereby constructing a technical closed loop of model building, residual correction, and performance verification, which ensures the reliability of the lateral force estimation results.

[0131] The tire lateral force estimation device provided by the present invention will be described below. The tire lateral force estimation device described below can be referred to in correspondence with the tire lateral force estimation method described above.

[0132] Figure 4 This is a schematic diagram of the tire lateral force estimation device provided by the present invention, as shown below. Figure 4 As shown, the tire lateral force estimation device includes: a parameter analysis module 410, an estimation module 420, a feature fusion module 430, and a lateral force calculation module 440.

[0133] The parameter analysis module 410 is used to optimize the parameters of the lateral force physical estimation model of the tire, obtain the key parameters of the model, and perform sensitivity analysis on the key parameters of the model to obtain the target stiffness parameters; the lateral force physical estimation model is used to characterize the correlation between the tire's acceleration time-domain signal and the lateral force. The estimation module 420 is used to determine the mapping relationship between the target stiffness parameter and the vertical load of the tire; the mapping relationship is used to update the parameters of the lateral force physical estimation model, and the time-domain acceleration signal of the tire is input into the updated lateral force physical estimation model to obtain the prior estimate of the lateral force. The feature fusion module 430 is used to extract time-domain features from the tire's ground contact time-domain signal and fuse the time-domain features with the prior estimate of the lateral force to obtain fused features; The lateral force calculation module 440 is used to perform nonlinear calculations on the fused features and determine the lateral force estimation result of the tire based on the nonlinear calculation results and the prior estimate of the lateral force.

[0134] The tire lateral force estimation device provided in this invention determines reliable physical benchmark parameters through model parameter optimization and selects core stiffness parameters of the model through sensitivity analysis, providing reliable data support for subsequent reduction of computational complexity and targeted high-precision parameter correction. By establishing and utilizing a load-stiffness adaptive correction mechanism, the key parameters of the physical benchmark model are dynamically adjusted according to the working conditions. By extracting the nonlinear dynamic characteristics of tire forces under transient working conditions and introducing prior physical data, a more robust fusion feature is obtained. The actual lateral force of the tire is comprehensively evaluated by combining nonlinear calculation results with prior lateral force estimates. By combining data-driven and model-driven approaches, the interpretability and accuracy of tire lateral force evaluation under complex working conditions are improved.

[0135] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5As shown, the electronic device may include: a processor 510, a communications interface 520, a memory 530, and a communications bus 540, wherein the processor 510, the communications interface 520, and the memory 530 communicate with each other through the communications bus 540. The processor 510 can call logic instructions in the memory 530 to execute a tire lateral force estimation method. This method includes: optimizing the parameters of a physical estimation model of the tire's lateral force to obtain key model parameters, and performing sensitivity analysis on these key parameters to obtain target stiffness parameters; the physical estimation model of the lateral force is used to characterize the correlation between the tire's acceleration time-domain signal and the lateral force; determining the mapping relationship between the target stiffness parameters and the tire's vertical load; updating the parameters of the physical estimation model of the lateral force using the mapping relationship, and inputting the tire's acceleration time-domain signal into the updated physical estimation model of the lateral force to obtain a priori estimates of the lateral force; extracting time-domain features from the tire's ground contact time-domain signal, and fusing the time-domain features with the priori estimates of the lateral force to obtain fused features; performing nonlinear calculations on the fused features, and determining the tire's lateral force estimation result based on the nonlinear calculation results and the priori estimates of the lateral force.

[0136] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0137] On the other hand, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the tire lateral force estimation method provided by the above methods. This method includes: optimizing the parameters of a physical estimation model of tire lateral force to obtain key model parameters, and performing sensitivity analysis on the key model parameters to obtain target stiffness parameters; the physical estimation model of lateral force is used to characterize the correlation between the tire's acceleration time-domain signal and lateral force; determining the mapping relationship between the target stiffness parameters and the tire's vertical load; updating the parameters of the physical estimation model of lateral force using the mapping relationship, and inputting the tire's acceleration time-domain signal into the updated physical estimation model of lateral force to obtain a priori estimates of lateral force; extracting time-domain features from the tire's ground contact time-domain signal, and fusing the time-domain features with the priori estimates of lateral force to obtain fused features; performing nonlinear calculations on the fused features, and determining the tire's lateral force estimation result based on the nonlinear calculation results and the priori estimates of lateral force.

[0138] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0139] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0140] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions 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 invention.

Claims

1. A method for estimating tire lateral force, characterized in that, include: The parameters of the physical estimation model of the lateral force of the tire are optimized to obtain the key parameters of the model, and the sensitivity analysis of the key parameters of the model is performed to obtain the target stiffness parameters. The physical estimation model for lateral force is used to characterize the correlation between the time-domain acceleration signal of the tire and the lateral force; Determine the mapping relationship between the target stiffness parameter and the vertical load of the tire; use the mapping relationship to update the parameters of the lateral force physical estimation model, and input the time-domain acceleration signal of the tire into the updated lateral force physical estimation model to obtain the prior estimate of the lateral force; The time-domain features are extracted from the tire's ground contact time-domain signal, and the time-domain features are fused with the prior estimate of the lateral force to obtain fused features; The fusion features are subjected to nonlinear calculations, and the lateral force estimation result of the tire is determined based on the nonlinear calculation results and the prior estimate of the lateral force.

2. The tire lateral force estimation method according to claim 1, characterized in that, The physical estimation model of the lateral force of the tire is optimized by parameters to obtain key model parameters, including: An improved sparrow search algorithm was used to optimize the parameters, and the key parameters of the obtained model were obtained. The improved sparrow search algorithm is used to initialize the position of individuals in the population using the Tent chaotic mapping during the iterative optimization process, and to introduce a golden sine mechanism in the position update strategy of the population discoverer to adjust the search step size; after each generation of iterative update, the crossover and mutation mechanism of the differential evolution algorithm is used to perturb the individuals in the population.

3. The tire lateral force estimation method according to claim 1, characterized in that, The sensitivity analysis of the key parameters of the model is performed to obtain the target stiffness parameters, including: The root mean square error of the model is calculated using the single-factor analysis method when the key parameters of the model change, and the fluctuation amplitudes corresponding to the root mean square errors of the model are calculated. Different fluctuation amplitudes correspond to different sidewall lateral stiffness. The lateral stiffness of the sidewall corresponding to the maximum value among the multiple fluctuation amplitudes is determined as the target stiffness parameter.

4. The tire lateral force estimation method according to claim 1, characterized in that, The step of extracting time-domain features from the tire's ground contact time-domain signal and fusing the time-domain features with the prior estimate of the lateral force to obtain fused features includes: The grounding time-domain signal is input into the lateral force residual correction model; the lateral force residual correction model includes a convolutional neural network branch and a long short-term memory network branch; The grounding time-domain signal is processed using the convolutional neural network branch to obtain local transient features; the grounding time-domain signal is processed using the long short-term memory network branch to obtain time-series features. The local transient features, the temporal features, and the prior estimate of the lateral force are concatenated along the feature dimension to generate the fused features.

5. The tire lateral force estimation method according to claim 4, characterized in that, The lateral force residual correction model also includes a multilayer perceptron residual mapping layer; The step of performing nonlinear calculations on the fused features and determining the lateral force estimation result of the tire based on the nonlinear calculation results and the prior estimate of the lateral force includes: The fused features are nonlinearly mapped through the multilayer perceptron residual mapping layer to obtain the lateral force residual correction amount; Using the lateral force residual correction amount as the nonlinear calculation result, the lateral force residual correction amount is added to the lateral force prior estimate to obtain the lateral force estimation result of the tire.

6. The tire lateral force estimation method according to any one of claims 4, characterized in that, The physical estimation model for the lateral force is constructed through the following steps: Modal expansion calculations are performed on the three-dimensional physical model of the tire to obtain a coefficient matrix, which is used to characterize the equivalent circumferential distribution. The acceleration time-domain signal of the tire is converted into an equivalent distributed excitation using the coefficient matrix. The equivalent distributed excitation is then solved in conjunction with the effective rolling radius of the tire. An analytical mapping relationship between the acceleration time-domain signal and the average lateral force within a single tire cycle is constructed to obtain the physical estimation model of the lateral force.

7. The tire lateral force estimation method according to any one of claims 4-6, characterized in that, The method further includes: Obtain the true value of the lateral force of the tire; Calculate the root mean square error, mean absolute error, normalized root mean square error, and coefficient of determination between the estimated lateral force of the tire and the true lateral force. Based on the calculated root mean square error, mean absolute error, normalized root mean square error, and coefficient of determination, the degree of deviation between the lateral force estimation result and the true value is evaluated, and model performance evaluation results are generated.

8. A tire lateral force estimation device, characterized in that, include: The parameter analysis module is used to optimize the parameters of the lateral force physical estimation model of the tire, obtain the key parameters of the model, and perform sensitivity analysis on the key parameters of the model to obtain the target stiffness parameters; the lateral force physical estimation model is used to characterize the correlation between the tire's acceleration time-domain signal and the lateral force. An estimation module is used to determine the mapping relationship between the target stiffness parameter and the vertical load of the tire; update the parameters of the lateral force physical estimation model using the mapping relationship, and input the time-domain acceleration signal of the tire into the updated lateral force physical estimation model to obtain the prior estimate of the lateral force. The feature fusion module is used to extract time-domain features from the tire's ground contact time-domain signal and fuse the time-domain features with the prior estimate of the lateral force to obtain fused features; The lateral force calculation module is used to perform nonlinear calculations on the fused features and determine the lateral force estimation result of the tire based on the nonlinear calculation results and the prior lateral force estimate.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the tire lateral force estimation method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the tire lateral force estimation method as described in any one of claims 1 to 7.

Images