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

Abstract

The invention provides a real-time wheel center force prediction method and device and a server, and relates to the technical field of vehicle dynamics modeling and intelligent control, and the method comprises the steps: obtaining high-latitude original feature data in real time, the original feature data comprises wheel center acceleration data and dynamics auxiliary data, and the wheel center acceleration data and the dynamics auxiliary data are obtained; the dynamic auxiliary data comprises the vehicle speed, the steering angle, the tire pressure, the tire force, the vehicle force and the vehicle dynamic relation; principal component analysis and physical constraint are combined, dynamic order reduction processing is carried out on the original feature data, and target feature data after order reduction are determined; and performing calculation processing on weight values corresponding to different input features in the target feature data through a preset deep neural network to dynamically adjust the importance of each input feature, and performing weighted calculation on the weight values and the target feature data to generate a target wheel center force prediction result in an output layer of the preset deep neural network. According to the method, the wheel center force prediction accuracy can be remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle dynamics modeling and intelligent control, and in particular, to a method, device and server for predicting real-time wheel center force. Background Art

[0002] The wheel center force is an important indicator for describing the dynamic operation of a vehicle. However, directly measuring the wheel center force by means of physical sensors has a high measurement cost and complex sensor installation problems. At present, related technologies propose that the wheel center force can be deduced by combining a physical model with vehicle structure parameters, but this solution has complex calculations and is difficult to adapt to different vehicles and working conditions. In addition, algorithms such as neural networks can be used to learn prediction models from data, but this solution is prone to overfitting or degradation of generalization performance when dealing with data with too high dimensions or non-stationary characteristics, thus affecting the accuracy of wheel center force prediction. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide a method, device and server for predicting real-time wheel center force, which can significantly improve the accuracy of wheel center force prediction.

[0004] In a first aspect, an embodiment of the present invention provides a method for predicting real-time wheel center force. The method includes: obtaining high-dimensional original feature data in real time, where the original feature data includes: wheel center acceleration data and dynamic auxiliary data, and the dynamic auxiliary data includes: vehicle speed, steering angle, tire pressure, tire force, vehicle body force and vehicle dynamics relationship; combining principal component analysis with physical constraints to perform dynamic dimensionality reduction processing on the original feature data to determine the target feature data after dimensionality reduction; calculating and processing the weight values corresponding to different input features in the target feature data through a preset deep neural network to dynamically adjust the importance of each input feature, and generating a target wheel center force prediction result at the output layer of the preset deep neural network by performing weighted calculation on the weight values and the target feature data.

[0005] In an implementation manner, the step of combining principal component analysis with physical constraints to perform dynamic dimensionality reduction processing on the original feature data to determine the target feature data after dimensionality reduction includes: performing dynamic dimensionality reduction processing on the original feature data through a preset principal component analysis model, mapping the original feature data from a high-dimensional space to a low-dimensional feature space, and making the reduced-dimensional feature space conform to the vehicle dynamics model by adding a first physical constraint term during the dynamic dimensionality reduction process to determine the target feature data after dimensionality reduction, where the objective function of the preset principal component analysis model is: Z = W^T X where W is the dimensionality reduction matrix, X is the original feature data, and Z is the target feature data in the low dimension after dimensionality reduction.

[0006] In one embodiment, the steps of making the feature space after dimensionality reduction conform to the vehicle dynamics model by adding a first physical constraint term during the dynamic dimensionality reduction process include: updating the objective function of the preset principal component analysis model through the first physical constraint term to determine the updated objective function; performing dynamic dimensionality reduction processing on the original feature data using the updated objective function to determine the target feature data after dimensionality reduction, where the updated objective function is:

[0007] where, is the first physical constraint term, is the weight coefficient for balancing the reconstruction error and physical consistency.

[0008] In one embodiment, the method includes: determining the first physical constraint term according to the constraint relationship between the tire force and the vehicle body force, where the first physical constraint term is expressed as:

[0009] where, is the tire force, is the friction coefficient, is the normal force, is the vehicle body force, is the vehicle body mass, is the vehicle body acceleration.

[0010] In one embodiment, after the step of determining the target feature data after dimensionality reduction, it includes: obtaining the real-time working condition of the vehicle and the data distribution of the target feature data, and using a preset dynamic optimization model to perform dynamic adjustment processing on the dimensions of various types of data in the target feature data according to the real-time working condition and the data distribution, so that in the optimized target feature data, the dimension of the data is proportional to the correlation degree predicted by the data and the wheel force.

[0011] In one embodiment, before the step of generating the target wheel force prediction result at the output layer of the preset deep neural network, it includes: determining a second physical constraint term using the relationship between force and vehicle dynamics, and determining a loss function based on the second physical constraint term to optimize the model parameters of the neural network, where the loss function is:

[0012] where, is the real wheel force, is the predicted wheel force, is the second physical constraint term.

[0013] In one embodiment, the step of determining the second physical constraint term by using the relationship between force and vehicle dynamics includes: determining the maximum allowable force of the vehicle under different working conditions based on the relationship between force and vehicle dynamics, and determining the second physical constraint term according to the maximum allowable force, where the second physical constraint term is:

[0014] where is the predicted value of the force, is the maximum allowable force of the vehicle under different working conditions.

[0015] In a second aspect, an embodiment of the present invention further provides a device for predicting real-time wheel center force. The device includes: a data acquisition module that acquires high-dimensional original feature data in real time, where the original feature data includes: wheel center acceleration data and dynamic auxiliary data, and the dynamic auxiliary data includes: vehicle speed, steering angle, tire pressure, tire force, vehicle body force, and vehicle dynamics relationship; a dynamic order reduction module that combines principal component analysis and physical constraints to perform dynamic order reduction processing on the original feature data to determine the target feature data after order reduction; a wheel center force prediction module that calculates and processes the weight values corresponding to different input features in the target feature data through a preset deep neural network to dynamically adjust the importance of each input feature, and generates a target wheel center force prediction result at the output layer of the preset deep neural network by performing weighted calculation on the weight values and the target feature data.

[0016] In a third aspect, an embodiment of the present invention further provides a server, including a processor and a memory. The memory stores computer-executable instructions that can be executed by the processor, and the processor executes the computer-executable instructions to implement the method according to any one of the first aspect.

[0017] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium. The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are called and executed by the processor, the computer-executable instructions cause the processor to implement the method according to any one of the first aspect.

[0018] The embodiments of the present invention bring the following beneficial effects: A method, apparatus, and server for predicting real-time wheel forces provided by an embodiment of the present invention. After obtaining high-dimensional original feature data in real time, the method combines principal component analysis with physical constraints to perform dynamic dimensionality reduction processing on the original feature data to determine the target feature data after dimensionality reduction. Then, through a preset deep neural network, the weight values corresponding to different input features in the target feature data are calculated and processed to dynamically adjust the importance of each input feature, and by performing weighted calculation on the weight values and the target feature data, a target wheel force prediction result is generated at the output layer of the preset deep neural network. The embodiment of the present invention can combine a neural network, a dimensionality reduction model, and an adaptive learning mechanism to predict wheel forces in real time and efficiently based on high-dimensional input data collected by vehicle sensors, and significantly improve the accuracy of wheel force prediction, providing support for intelligent driving decision-making, vehicle dynamic control, and tire performance optimization.

[0019] Other features and advantages of the present invention will be described in the following specification, and, in part, will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention are achieved and obtained by the structures specifically pointed out in the specification, claims, and drawings.

[0020] To make the above objectives, features, and advantages of the present invention more obvious and understandable, the following specifically enumerates preferred embodiments and, in conjunction with the accompanying drawings, provides a detailed description as follows. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0022] Figure 1 It is a schematic flowchart of a method for predicting real-time wheel forces provided by an embodiment of the present invention; Figure 2 It is a schematic flowchart of another method for predicting real-time wheel forces provided by an embodiment of the present invention; Figure 3 It is a schematic structural diagram of a device for predicting real-time wheel forces provided by an embodiment of the present invention; Figure 4 It is a schematic structural diagram of a server provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0024] At present, wheel center force is an important indicator for describing the dynamic operation of a vehicle, including longitudinal force (Fx), lateral force (Fy), and vertical force (Fz), which can characterize the traction, braking, and lateral stability performance of the vehicle. However, directly measuring wheel center force through physical sensors has a high measurement cost and complex sensor installation problems. Related technologies propose that based on the physical model method, wheel center force can be deduced through dynamic formulas and vehicle structure parameters, but this solution has complex calculations and is difficult to adapt to different vehicles and working conditions. In addition, based on the data-driven method, algorithms such as neural networks can be used to learn prediction models from data, but this solution is prone to overfitting or a decline in generalization performance when dealing with data with too high dimensions or non-stationary characteristics, thus affecting the accuracy of wheel center force prediction.

[0025] Based on this, the real-time wheel center force prediction method, device, and server provided by the embodiments of the present invention can significantly improve the prediction accuracy and real-time performance under complex working conditions. Through dynamic order reduction modeling, high-dimensional data redundancy can be effectively reduced, key features can be retained, and the model complexity can be reduced. In addition, multi-modal data fusion and physical constraint optimization ensure the physical consistency of the prediction results and solve the instability problem of traditional methods. Combining the non-linear modeling and adaptive parameter adjustment mechanism of neural networks, the model can adapt to different road surfaces and driving conditions in real time, achieving high-precision prediction of longitudinal force, lateral force, and vertical force. Compared with traditional methods, the prediction accuracy is increased by more than 15%, and the calculation efficiency is increased by 30%. It is widely applicable to the fields of intelligent vehicle control and autonomous driving.

[0026] See Figure 1 The flowchart of a real-time wheel center force prediction method shown, this method mainly includes the following steps S102 to step S106: Step S102, obtain high-dimensional original feature data in real time, where the original feature data includes: wheel center acceleration data and dynamic auxiliary data, and the dynamic auxiliary data includes: vehicle speed, steering angle, tire pressure, tire force, vehicle body force, and vehicle dynamics relationship.

[0027] Step S104: Combine principal component analysis with physical constraints to perform dynamic dimensionality reduction on the original feature data, and determine the target feature data after dimensionality reduction. In one implementation, the dimensionality reduction dimension can be dynamically adjusted, and the number of principal components can be dynamically determined according to the real-time working conditions and data distribution to adapt to the non-stationary characteristics of the input data.

[0028] Step S106: Calculate and process the weight values corresponding to different input features in the target feature data through a preset deep neural network to dynamically adjust the importance of each input feature, and generate a target wheel force prediction result at the output layer of the preset deep neural network by performing weighted calculation on the weight values and the target feature data. Among them, the deep neural network (DNN) includes: an input layer, a hidden layer, an attention mechanism module, and an output layer. Specifically, the input layer is used to receive the low-dimensional feature vector Z after dimensionality reduction (i.e., the target feature data); the hidden layer consists of multiple fully connected networks, and the activation function of each layer uses ReLU: f(x)=max(0,x); the attention mechanism module is used to dynamically adjust the importance of the input features by calculating the weights of the features, help the network pay more attention to the key features, improve the expression ability and accuracy of the model, and multiply the weight values of different features by the input data through a weighted sum mechanism so that the neural network can focus on the features more important for wheel force prediction; the output layer is used to predict the longitudinal force Fx, the lateral force Fy, and the vertical force Fz (i.e., the target wheel force prediction result).

[0029] The above real-time wheel force prediction method provided by the embodiments of the present invention can significantly improve the accuracy of wheel force prediction.

[0030] The embodiments of the present invention also provide an implementation manner that reduces high-dimensional data redundancy, retains key features, and reduces the model complexity through dynamic dimensionality reduction modeling when predicting wheel forces in real time. Specifically, refer to the following (1) to (3): (1) Adopt a dynamic dimensionality reduction method that combines principal component analysis (PCA) with physical constraints. The role of PCA is to reduce the dimension by retaining the main components of the data while retaining as much original information as possible, and map the data from a high-dimensional space to a low-dimensional feature space. Specifically, through a preset principal component analysis model, perform dynamic dimensionality reduction processing on the original feature data, map the original feature data from a high-dimensional space to a low-dimensional feature space, and add a first physical constraint term during the dynamic dimensionality reduction process to make the reduced-dimensional feature space conform to the vehicle dynamics model to determine the target feature data after dimensionality reduction. Among them, the objective function of the preset principal component analysis model is: Z = W^T X where W is the dimensionality reduction matrix, X is the original feature data, and Z is the low-dimensional target feature data after dimensionality reduction.

[0031] (2) During the dimension reduction process, a physical constraint term is added to ensure that the feature space after dimension reduction conforms to the vehicle dynamics model. Specifically, the objective function of the preset principal component analysis model is updated through the first physical constraint term to determine the updated objective function; The original feature data is dynamically dimensionally reduced using the updated objective function to determine the target feature data after dimension reduction, where the updated objective function is:

[0032] Among them, is the first physical constraint term, is the weight coefficient for balancing the reconstruction error and physical consistency.

[0033] Furthermore, according to the constraint relationship between the tire force and the vehicle body force, the first physical constraint term is determined, where the first physical constraint term is expressed as:

[0034] Among them, is the tire force, is the friction coefficient, is the normal force, is the vehicle body force, is the vehicle body mass, is the vehicle body acceleration.

[0035] (3) According to the real-time working conditions and data distribution, the dimension reduction dimension k is dynamically adjusted to ensure that the features after dimension reduction retain as much important information as possible. Specifically, the real-time working conditions of the vehicle and the data distribution of the target feature data are obtained, and using the preset dynamic optimization model, according to the real-time working conditions and data distribution, the dimensions of various types of data in the target feature data are dynamically adjusted so that in the optimized target feature data, the dimensions of the data are proportional to the correlation degree with the data and the wheel force prediction. The dynamic adjustment of the dimension reduction dimension can be achieved through the following formula:

[0036] Among them, is the eigenvalue of the covariance matrix; τ is the cumulative variance contribution rate threshold, with a value of 95%.

[0037] The embodiment of the present invention also provides an implementation method for predicting wheel forces in real time, which combines the nonlinear modeling and adaptive parameter adjustment mechanism of a neural network. The model can adapt to different road surfaces and driving conditions in real time and achieve high-precision prediction of longitudinal force, lateral force, and vertical force. Specifically, refer to the following (1) to (2): (1) To ensure the accuracy of the prediction results, physical constraints are introduced and a loss function is defined. Specifically, the second physical constraint term is determined using the relationship between force and vehicle dynamics, and the loss function is determined based on the second physical constraint term to optimize the model parameters of the neural network using the loss function. Among them, the loss function is:

[0038] Among them, is the true wheel force, is the predicted wheel force, is the second physical constraint term.

[0039] Based on the relationship between force and vehicle dynamics, the maximum allowable force of the vehicle under different working conditions is determined, and the second physical constraint term is determined according to the maximum allowable force. Among them, the second physical constraint term is:

[0040] Among them, is the predicted value of the force, is the maximum allowable force of the vehicle under different working conditions.

[0041] (2) Under different working conditions, in order to improve the adaptability of the model, the Bayesian optimization method can be used to dynamically adjust the key parameters of the model, so as to avoid overfitting or underfitting problems while optimizing the model performance. To improve the adaptability of the model under different working conditions, the Bayesian optimization is used to dynamically adjust the key parameters of the model (such as the learning rate, regularization coefficient, etc.). Among them, the surrogate model of the objective function can be constructed through the Gaussian process, and the acquisition function is defined as the expected improvement (EI):

[0042] Thus, by maximizing the best parameter update direction is found.

[0043] In one implementation, the dynamic parameter update process includes: initialization, real-time working condition monitoring, Bayesian search, and parameter update. Specifically, in the initialization process, the model is trained using the initial parameters and the loss value is recorded. In the real-time working condition monitoring process, the parameter optimization is triggered. In the Bayesian search process, the optimal parameters are found. In the parameter update process, the model parameters are adjusted and the training continues, so as to efficiently search for the global optimal parameters and improve the prediction accuracy and model robustness under complex dynamic working conditions.

[0044] See Figure 2Flow schematic diagram of another real-time wheel force prediction method. In modern automotive engineering, real-time prediction of wheel forces (including longitudinal force, lateral force, and vertical force) is crucial for enhancing vehicle handling, safety, and comfort. Especially under different road conditions and driving scenarios, accurate prediction of wheel forces can assist the control system in real-time adjusting the vehicle state and improving the driving experience. This example demonstrates how to use a prediction method based on the fusion of dynamic order reduction and neural network, combined with sensor data and vehicle dynamics models, to predict wheel forces.

[0045] 1. Input Feature Selection and Data Acquisition 1.1 Wheel Center Acceleration Data Three-axis acceleration data, including longitudinal acceleration, lateral acceleration, and vertical acceleration, is collected in real-time through accelerometers on each wheel of the vehicle. Assume that each wheel is equipped with an acceleration sensor, and the data collection frequency is 100Hz.

[0046] The wheel center acceleration data includes the following, represented as a time series: Time t1: Longitudinal acceleration ax = 1.2m / s2, Lateral acceleration ay = 0.5m / s2, Vertical acceleration az = -9.8m / s2.

[0047] Time t2: Longitudinal acceleration ax = 1.3m / s2, Lateral acceleration ay = 0.6m / s2, Vertical acceleration az = -9.7m / s2.

[0048] 1.2 Kinematic Auxiliary Data: Vehicle speed: Collected through a vehicle speed sensor. Assume the vehicle speed data is as follows: Time t1: Vehicle speed v = 80km / h.

[0049] Time t2: Vehicle speed v = 85km / h.

[0050] Steering angle: Obtained from a steering angle sensor. Assume the steering angle data is as follows: Time t1: Steering angle δ = 2°. Time t2: Steering angle δ = 3°. Tire pressure: The pressure of each tire is provided in real-time by a tire pressure sensor. Assume the tire pressure data is as follows: Time t1: Tire pressure p = 2.3bar.

[0051] Time t2: Tire pressure p = 2.2bar.

[0052] Road surface condition: The road surface friction coefficient is provided in real-time by a ground friction sensor. Assume the friction coefficient is μ = 0.9 at time t1 and t2.

[0053] The above data is synchronously collected through the vehicle's control system (such as an on-vehicle computer) and transmitted to the wheel force prediction system in real time.

[0054] 2. Data Preprocessing and Dynamic Dimensionality Reduction 2.1 Data Preprocessing: Denoising: Sensor data is denoised through Kalman filtering to filter out high-frequency noise and ensure data quality.

[0055] Normalization: All input data is normalized so that its mean is 0 and variance is 1 to ensure consistency when the data is input into the neural network.

[0056] 2.2 Dynamic Dimensionality Reduction Processing: PCA Dimensionality Reduction: First, perform PCA dimensionality reduction on all the above input features to obtain a low-dimensional feature space.

[0057] Introduce Physical Constraints: During the dimensionality reduction process, add physical constraints of vehicle dynamics to ensure that the reduced features satisfy the vehicle mechanics model. For example, by calculating the constraint relationship between tire forces and vehicle body forces, ensure the physical consistency of the data in the reduced feature space.

[0058] The physical constraint terms are as follows:

[0059] Among them, is the tire force, is the friction coefficient, is the normal force, is the vehicle body force, is the vehicle body mass.

[0060] is the vehicle body acceleration.

[0061] Dynamically Adjust the Dimensionality Reduction Dimension: According to the real-time working conditions and data distribution, dynamically adjust the dimensionality reduction dimension k to ensure that the reduced features retain the most important information.

[0062] 3. Adaptive Neural Network Modeling 3.1 Neural Network Architecture: Input Layer: The reduced features Z are used as the input, with a size of N×k, where N is the number of samples and k is the dimension of the reduced features.

[0063] Hidden Layer: A multi-layer fully connected network is adopted, and each layer uses the ReLU activation function for non-linear mapping.

[0064] · Hidden Layer 1: 512 nodes, activation function ReLU · Hidden Layer 2: 256 nodes, activation function ReLU · Hidden layer 3: 128 nodes, activation function ReLU Attention mechanism module: This module is used to dynamically adjust the weights of input features, calculate the weights of each feature based on the weighted sum mechanism, and then perform weighted calculation. This module helps the network focus on the most important features and improve the accuracy of prediction.

[0065] Output layer: Output the three wheel force components of the vehicle: · Longitudinal force Fx · Lateral force Fy · Vertical force Fz The unit of the output is Newton (N).

[0066] 4. Physical constraints and loss function 4.1 Loss function: To ensure the accuracy of the prediction results and comply with physical laws, the loss function is designed as:

[0067] Where: is the true wheel force, is the predicted wheel force, is the physical constraint term, ensuring that the predicted force value conforms to the vehicle dynamics constraints.

[0068] 4.2 Physical constraint term: The physical constraint term is used to constrain the relationship between the prediction results and the actual vehicle dynamics. The specific form is:

[0069] Where is the predicted force value, is the maximum allowable force of the vehicle under specific working conditions.

[0070] 5. Bayesian optimization and adaptive learning 5.1 Principle of Bayesian optimization: Objective: Dynamically adjust the hyperparameters in the neural network, such as the learning rate, regularization coefficient, etc., through Bayesian optimization, optimize the training process, and avoid overfitting or underfitting.

[0071] Optimization process: Use the Gaussian process model to construct a surrogate model of the objective function Define the acquisition function and maximize EI to search for the optimal hyperparameters.

[0072] 5.2 Dynamic parameter update: Real-time operating condition monitoring: Monitor the changes in the vehicle operating condition during each data input, and trigger the Bayesian optimization algorithm for hyperparameter optimization.

[0073] Optimization and update: Adjust the model hyperparameters such as the learning rate and regularization coefficient according to the current operating condition and loss function, so as to improve the prediction accuracy of the model under the new operating condition.

[0074] 6. Output results Prediction results: Time t1: · Longitudinal force Fx = 1200 · Lateral force Fy = 400N · Vertical force Fz = 8500N Time t2: · Longitudinal force Fx = 1250N · Lateral force Fy = 430N · Vertical force Fz = 8600N These predicted values are consistent with the actual operating condition and physical constraints of the vehicle, and can be effectively used to dynamically adjust the vehicle control system to ensure stability and safety during driving.

[0075] In summary, the present invention can improve the physical interpretability of feature selection by combining the improved dynamic reduced-order model with physical characteristics and principal component analysis (PCA) and integrating vehicle dynamics constraints during the dimensionality reduction process. In addition, by designing a hybrid neural network architecture with adaptive learning ability, using an online update mechanism to improve the adaptability of the model to operating condition changes, and using multi-modal fusion technology to combine sensor data with vehicle historical operating state data, accurate modeling of complex non-linear relationships can be achieved.

[0076] For the real-time wheel force prediction method provided in the foregoing embodiments, the embodiments of the present invention provide a real-time wheel force prediction device. Refer to Figure 3 The structural schematic diagram of a real-time wheel force prediction device shown, the device includes the following parts: Data acquisition module 302, which acquires high-dimensional original feature data in real time. Among them, the original feature data includes: wheel center acceleration data and dynamic auxiliary data. The dynamic auxiliary data includes: vehicle speed, steering angle, tire pressure, tire force, vehicle body force, and vehicle dynamics relationship; Dynamic reduced-order module 304, which combines principal component analysis with physical constraints to perform dynamic reduced-order processing on the original feature data to determine the target feature data after reduction; The wheel force prediction module 306 calculates and processes the weight values corresponding to different input features in the target feature data through a preset deep neural network, so as to dynamically adjust the importance of each input feature, and generates a target wheel force prediction result at the output layer of the preset deep neural network by performing weighted calculation on the weight values and the target feature data.

[0077] The above real-time wheel force prediction device provided by the embodiments of the present application can significantly improve the accuracy of wheel force prediction.

[0078] In one implementation manner, when performing the step of combining principal component analysis with physical constraints to perform dynamic dimensionality reduction processing on the original feature data to determine the target feature data after dimensionality reduction, the above dynamic dimensionality reduction module 304 is further configured to: perform dynamic dimensionality reduction processing on the original feature data through a preset principal component analysis model, map the original feature data from a high-dimensional space to a low-dimensional feature space, and make the feature space after dimensionality reduction conform to the vehicle dynamics model by adding a first physical constraint term during the dynamic dimensionality reduction process, so as to determine the target feature data after dimensionality reduction, where the objective function of the preset principal component analysis model is: Z = W^T X where W is the dimensionality reduction matrix, X is the original feature data, and Z is the low-dimensional target feature data after dimensionality reduction.

[0079] In one implementation manner, when performing the step of making the feature space after dimensionality reduction conform to the vehicle dynamics model by adding a first physical constraint term during the dynamic dimensionality reduction process, the above dynamic dimensionality reduction module 304 is further configured to: update the objective function of the preset principal component analysis model through the first physical constraint term to determine the updated objective function; perform dynamic dimensionality reduction processing on the original feature data by using the updated objective function to determine the target feature data after dimensionality reduction, where the updated objective function is:

[0080] where is the first physical constraint term, is the weight coefficient for balancing the reconstruction error and physical consistency.

[0081] In one implementation manner, the above dynamic dimensionality reduction module 304 is further configured to: determine the first physical constraint term according to the constraint relationship between the tire force and the vehicle body force, where the first physical constraint term is expressed as:

[0082] where is the tire force, is the friction coefficient, is the normal force, is the vehicle body force, is the vehicle body mass, is the vehicle body acceleration.

[0083] In one implementation, after the step of determining the target feature data after reduction, the above dynamic reduction module 304 is further configured to: obtain the real-time working condition of the vehicle and the data distribution of the target feature data, and use a preset dynamic optimization model to dynamically adjust the dimensions of various types of data in the target feature data according to the real-time working condition and the data distribution, so that in the optimized target feature data, the dimension of the data is proportional to the degree of correlation with the wheel force prediction.

[0084] In one implementation, before the step of generating the target wheel force prediction result by the output layer of the preset deep neural network, it includes: determining a second physical constraint term using the relationship between force and vehicle dynamics, and determining a loss function based on the second physical constraint term to optimize the model parameters of the neural network using the loss function, where the loss function is:

[0085] where, is the real wheel force, is the predicted wheel force, is the second physical constraint term.

[0086] In one implementation, the step of determining the second physical constraint term using the relationship between force and vehicle dynamics includes: determining the maximum allowable force of the vehicle under different working conditions based on the relationship between force and vehicle dynamics, and determining the second physical constraint term according to the maximum allowable force, where the second physical constraint term is:

[0087] where, is the predicted value of the force, is the maximum allowable force of the vehicle under different working conditions.

[0088] The device provided by the embodiments of the present invention has the same implementation principle and the same technical effects as those of the foregoing method embodiments. For a brief description, for the parts not mentioned in the device embodiments, reference may be made to the corresponding contents in the foregoing method embodiments.

[0089] The embodiments of the present invention provide a server. Specifically, the server includes a processor and a storage device; a computer program is stored on the storage device, and when the computer program is run by the processor, it executes the method according to any one of the above-mentioned implementations.

[0090] Figure 4A schematic structural diagram of a server provided by an embodiment of the present invention. The server 100 includes: a processor 40, a memory 41, a bus 42, and a communication interface 43. The processor 40, the communication interface 43, and the memory 41 are connected through the bus 42. The processor 40 is configured to execute an executable module stored in the memory 41, such as a computer program.

[0091] Among them, the memory 41 may include a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk memory. Through at least one communication interface 43 (which can be wired or wireless), a communication connection is established between this system network element and at least one other network element, and the Internet, wide area network, local area network, metropolitan area network, etc. can be used.

[0092] The bus 42 can be an ISA bus, a PCI bus, an EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience of representation, Figure 4 only a bidirectional arrow is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0093] Among them, the memory 41 is used to store a program. After receiving an execution instruction, the processor 40 executes the program. The method executed by the device defined by the flow process disclosed in any one of the foregoing embodiments of the present invention can be applied to the processor 40 or implemented by the processor 40.

[0094] The processor 40 may be an integrated circuit chip with the ability to process signals. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware in the processor 40 or the instructions in the form of software. The above-mentioned processor 40 may be a general-purpose processor, including a central processing unit (CPU for short), a network processor (NP for short), etc.; it may also be a digital signal processor (DSP for short), an application specific integrated circuit (ASIC for short), a field-programmable gate array (FPGA for short) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. It can implement or execute the various methods, steps and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in combination with the embodiments of the present invention can be directly embodied as being executed and completed by the hardware decoding processor, or executed and completed by the combination of the hardware and software modules in the decoding processor. The software module may be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. This storage medium is located in the memory 41, and the processor 40 reads the information in the memory 41 and combines its hardware to complete the steps of the above method.

[0095] The computer program product of the readable storage medium provided by the embodiments of the present invention includes a computer-readable storage medium storing program code, and the instructions included in the program code can be used to execute the method described in the foregoing method embodiments. For the specific implementation, reference can be made to the foregoing method embodiments and will not be elaborated here.

[0096] When the above-mentioned function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, 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. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROMs, Read-Only Memories), random access memories (RAMs, Random Access Memories), magnetic disks, or optical discs that can store program codes.

[0097] Finally, it should be noted that the above-mentioned embodiments are only specific embodiments of the present invention, used to illustrate the technical solutions of the present invention, rather than limiting it. The protection scope of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that any person skilled in the art within the technical scope disclosed by the present invention can still modify the technical solutions recorded in the foregoing embodiments or easily conceive of changes, or perform equivalent replacements for some of the technical features; and these modifications, changes, or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention and should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A real-time wheel force prediction method, characterized in that: The method comprises: Acquire high-dimensional original feature data in real time, wherein the original feature data includes: wheel center acceleration data and dynamic auxiliary data, and the dynamic auxiliary data includes: vehicle speed, steering angle, tire pressure, tire force, vehicle body force and vehicle dynamics relationship; Combining principal component analysis with physical constraints, dynamically reducing the original feature data, and determining the target feature data after the reduction; The weight values ​​corresponding to different input features in the target feature data are calculated and processed through a preset deep neural network to dynamically adjust the importance of each input feature, and the weight values ​​and the target feature data are weightedly calculated to generate a target wheel centrifugal force prediction result in the output layer of the preset deep neural network.

2. The real-time wheel center force prediction method according to claim 1, characterized in that: The step of combining principal component analysis with physical constraints, dynamically reducing the original feature data, and determining the target feature data after the reduction includes: The original feature data is dynamically reduced in order by using a preset principal component analysis model, the original feature data is mapped from a high-dimensional space to a low-dimensional feature space, and the first physical constraint term is added in the dynamic reduction process so that the reduced-dimensional feature space conforms to the vehicle dynamics model to determine the target feature data after the reduced-dimensionality, wherein the objective function of the preset principal component analysis model is: Z=WTX Among them, W is the reduced-order matrix, X is the original feature data, and Z is the low-dimensional target feature data after dimensionality reduction.

3. The real-time wheel center force prediction method according to claim 2, characterized in that: The steps of making the feature space after dimension reduction conform to the vehicle dynamics model by adding the first physical constraint term in the dynamic order reduction process include: Updating the objective function of the preset principal component analysis model by using the first physical constraint item to determine an updated objective function; The original feature data is dynamically reduced in order using the updated objective function to determine the target feature data after dimension reduction, wherein the updated objective function is: in, is the first physical constraint, is the weight coefficient for balancing reconstruction error and physical consistency.

4. The real-time wheel center force prediction method according to claim 3, characterized in that: The method comprises: The first physical constraint item is determined according to the constraint relationship between the tire force and the vehicle body force, wherein the first physical constraint item is expressed as: in, is the tire force, is the friction coefficient, is the normal force, For the car's physical strength, is the vehicle body mass, is the vehicle acceleration.

5. The real-time wheel center force prediction method according to claim 2, characterized in that: After the step of determining the reduced-order target feature data, the method further comprises: The real-time working condition of the vehicle and the data distribution of the target feature data are obtained, and a preset dynamic optimization model is used to dynamically adjust the dimensions of each type of data in the target feature data according to the real-time working condition and the data distribution, so that in the optimized target feature data, the dimension of the data is proportional to the degree of correlation between the data and the wheel center force prediction.

6. The real-time wheel center force prediction method according to claim 1, characterized in that: Before the step of generating the target wheel centrifugal force prediction result at the output layer of the preset deep neural network, the method includes: A second physical constraint term is determined by using the relationship between the force and the vehicle dynamics, and a loss function is determined based on the second physical constraint term, so as to optimize the model parameters of the neural network by using the loss function, wherein the loss function is: in, For the real wheel of mind power, To predict the wheel force, is the second physical constraint.

7. The real-time wheel center force prediction method according to claim 6, characterized in that: The step of determining the second physical constraint term by using the relationship between the force and the vehicle dynamics comprises: Based on the relationship between force and vehicle dynamics, the maximum allowable force of the vehicle under different working conditions is determined, and the second physical constraint item is determined according to the maximum allowable force, wherein the second physical constraint item is: in, is the predicted value of force, It is the maximum allowable force of the vehicle under different working conditions.

8. A real-time wheel force prediction device, characterized in that: The device comprises: A data acquisition module acquires high-dimensional original feature data in real time, wherein the original feature data includes wheel center acceleration data and dynamic auxiliary data, and the dynamic auxiliary data includes vehicle speed, steering angle, tire pressure, tire force, vehicle body force and vehicle dynamics relationship; A dynamic order reduction module combines principal component analysis with physical constraints to dynamically reduce the original feature data and determine the target feature data after reduction; The wheel center force prediction module calculates and processes the weight values ​​corresponding to different input features in the target feature data through a preset deep neural network to dynamically adjust the importance of each input feature, and performs weighted calculation on the weight value and the target feature data to generate a target wheel center force prediction result in the output layer of the preset deep neural network.

9. A server, characterized in that: The method comprises a processor and a memory, wherein the memory stores computer executable instructions that can be executed by the processor, and the processor executes the computer executable instructions to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are called and executed by a processor, the computer-executable instructions prompt the processor to implement the method according to any one of claims 1 to 7.

Images