A vehicle wheel hub load online estimation method

Abstract

The present application relates to a kind of vehicle wheel core load online estimation method, electronic equipment and storage medium, the method includes the following steps: S1, using multitask Lasso and stability selection model to select vehicle sensor signal, obtain multiple leading sensor signal;S2, construct non-causal MIMO-FIR model;S3, based on principal component analysis PCA, short-time Fourier transform STFT and non-negative matrix factorization NMF extraction working condition feature;S4, based on extreme value machine EVM identification working condition;S5, according to working condition identification result, update non-causal MIMO-FIR model parameter, and output predicted vehicle wheel core load.The present application is relative to prior art in generalization, self-adaptability, real-time, five dimensions of explainability and engineering deployability, and has achieved systematic improvement, provides robust and economical soft measurement scheme for the chassis perception of next-generation intelligent vehicle.

Description

Technical Field

[0001] This invention relates to the field of vehicle measurement technology, and in particular to an online estimation method for vehicle wheel center load, electronic equipment, and storage medium. Background Technology

[0002] Wheel center load is a key dynamic quantity connecting the wheel and the vehicle body, and its accurate acquisition is crucial for applications such as electronic stability control, anti-lock braking, torque vectoring, active chassis control, and autonomous driving trajectory planning. In advanced vehicle dynamics control, active safety systems, and autonomous driving applications, accurate acquisition of tire-road contact force is fundamental to key functions such as stability control, anti-lock braking, and trajectory planning. While traditional hardware measurement methods—wheel force sensors—offer high accuracy, they are expensive, structurally fragile, significantly increase unsprung mass, and alter overall vehicle dynamics, making mass production and vehicle application difficult. To replace wheel force sensors, existing soft measurement methods are mainly divided into model-based observer methods and data-driven machine learning methods.

[0003] Model-based observer methods, represented by extended Kalman filtering, unscented Kalman filtering, and sliding mode observers, recursively estimate tire-road interaction forces using vehicle dynamics equations and measurable signals such as acceleration, angular velocity, and steering angle. While these methods offer clear physical meaning, they are highly dependent on modeling tire-road friction nonlinearities, variations in road adhesion coefficients, and suspension nonlinearities. Furthermore, they require linearization approximations and have limited ability to handle unknowns in road inputs.

[0004] Data-driven machine learning methods, such as neural networks, support vector regression, and fuzzy logic, learn input-output mappings from large-scale sensor data, enabling them to capture complex nonlinear relationships. However, these methods are generally limited by the closed-set assumption, meaning they can only be effectively estimated under conditions covered by the training data. When new road surfaces, new driving modes, or changes in vehicle configuration occur outside the training set, their generalization performance significantly decreases or even fails catastrophically.

[0005] The aforementioned soft measurement methods have the following three core problems that urgently need to be solved: First, there is the ill-conditioned inverse problem caused by unknown road excitation. The whole vehicle system can be regarded as a complex filter. The road surface roughness, as the system input, directly determines the wheel center load response. However, the road surface profile cannot be directly measured in production vehicles, which makes it a typical underdetermined and ill-conditioned inverse problem to infer the load from the measurable output. Traditional methods based on Kalman filtering or sliding mode observers must rely on linearization approximation and known road surface assumptions. Once encountering unmodeled road excitation, the accuracy degrades significantly.

[0006] Second, catastrophic generalization failure under open-set conditions. Traditional soft measurement methods generally assume closed-set conditions, that is, the training set covers all possible road surfaces and driving scenarios. Therefore, once encountering new road surfaces outside the training set (such as unseen rough road sections, speed bumps, potholes), violent maneuvers, or changes in vehicle configuration (such as load changes, tire changes, suspension modifications), the static model will experience unpredictable generalization failure, and the estimation error will be significantly amplified, seriously threatening the reliability of vehicle safety control.

[0007] Third, the continuous drift of model parameters under non-smooth driving environments. During the vehicle's life cycle, changes in load, tire wear, suspension aging, and hardware reconfiguration are inevitable. These gradual and abrupt changes will alter the system dynamics. However, current soft sensors, after offline training, are difficult to adaptively track these changes online. They lack the ability to simultaneously identify and respond to "gradual changes in known operating conditions" and "abrupt changes in unknown operating conditions." Furthermore, online updates based on gradients or heuristic algorithms are computationally expensive and difficult to deploy in automotive-grade edge computing units. Summary of the Invention

[0008] The purpose of this invention is to provide an online estimation method, electronic device, and storage medium for vehicle wheel center loads to solve the aforementioned technical problems. Therefore, the technical solution adopted by this invention is as follows: According to one aspect of the present invention, a method for online estimation of vehicle wheel center load is provided, comprising the following steps: S1. The vehicle sensor signals are selected using a multi-task Lasso and stability selection model to obtain multiple dominant sensor signals. S2. Construct a non-causal MIMO-FIR model; S3. Extracting working condition features based on principal component analysis (PCA), short-time Fourier transform (STFT), and nonnegative matrix factorization (NMF); S4. Working conditions are identified based on the extreme value machine EVM; S5. Based on the working condition identification results, update the parameters of the non-causal MIMO-FIR model and output the predicted vehicle wheel center load.

[0009] According to another aspect of the present invention, an electronic device is also provided, including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements the steps of the method described above.

[0010] According to another aspect of the present invention, a computer-readable storage medium is also provided, on which a computer program is stored, wherein, when the computer program is executed by a processor, it implements the steps of the method described above.

[0011] The present invention, by adopting the above technical solution, has the following advantages: Starting from the causal chain of "unknown road surface excitation → ill-conditioned inverse problem," this invention establishes a pure output-output mapping using a transitivity operator. This eliminates the dependence of wheel center load estimation on road surface profile measurements, fundamentally removing the existing observer's reliance on road surface priors. Therefore, it maintains a stable mapping on any unknown road surface, significantly improving cross-road surface generalization capability. Furthermore, the non-causal FIR structure simultaneously incorporates pseudo-inputs from finite past and future moments, enabling the model to precisely capture time delays and feedforward responses in vehicle dynamics. Compared to structures containing only causal terms, it performs better under transient conditions.

[0012] To address the problem of "closed set assumption → failure under new operating conditions," this invention introduces EVM to construct open set gating and uses Weibull distribution to fit the compact inclusion probability boundary of each class, enabling the system to reject "unseen operating conditions." The causal consequence is that the model no longer makes confident but incorrect estimates of unfamiliar inputs, but instead identifies them as new operating conditions and triggers the initialization and learning of the new model, thereby avoiding the catastrophic generalization failure that is common in existing static models and providing a reliable confidence signal for the upper-level control system.

[0013] To address the problem of "blind or unconditional online updates leading to degradation of steady-state accuracy," the condition-triggered update strategy of this invention updates the corresponding model only when a known operating condition is identified. This avoids noise injection under known operating conditions and also avoids forced correction of erroneous models under new operating conditions. The causal consequence is that the estimation is more stable under steady-state operating conditions and the response is faster under transient new operating conditions, achieving refined adaptation of "updating when it should be updated and maintaining when it should be maintained."

[0014] To address the issue of "channel redundancy → dimensionality and computational pressure," this invention obtains a compact subset of inputs shared across outputs through multi-task Lasso stability selection. The causal consequences are: significant compression of model parameter dimensions, improved identifiability, stable variable selection across multiple trials, and direct reduction of inference latency on edge devices. This allows the entire processing pipeline to be completed in an average of 4.8ms / cycle, meeting the 10ms automotive-grade real-time budget.

[0015] To address the issue of "offline training → inability to adapt to configuration changes," the RLS incremental update of this invention, combined with the Woodbury identity, enables the system to continuously update online without storing all historical data. When the load, tires, or suspension change, conditional gating automatically triggers the update, and the system can recover to near-baseline estimation accuracy within 5 to 15 seconds, providing out-of-the-box full lifecycle adaptive capability.

[0016] Finally, the edge-cloud collaborative architecture enables new operating condition models learned on prototype vehicles to be directly pushed to a fleet of vehicles of the same type. The causal consequence is that group-level learning accelerates model maturity, and individual vehicles do not need to learn all operating conditions independently to enjoy the experience "from other vehicles", which significantly shortens the system deployment and optimization cycle.

[0017] In summary, this invention achieves a systematic improvement over existing technologies in five dimensions: generalization, adaptability, real-time performance, interpretability, and engineering deployability, providing a robust and economical soft measurement solution for chassis perception in next-generation intelligent vehicles. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating an online estimation method for vehicle wheel center load according to an embodiment of the present invention. Figure 2 This is a flowchart of the working condition feature extraction process; Figure 3 This is a schematic diagram of the probability distribution of signals under different operating conditions. Detailed Implementation

[0019] The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings to provide a clearer understanding of the purpose, features, and advantages of the present invention. It should be understood that the embodiments shown in the drawings are not intended to limit the scope of the present invention, but are merely illustrative of the essential spirit of the technical solution of the present invention.

[0020] In the following description, certain specific details are set forth for the purpose of illustrating various disclosed embodiments in order to provide a thorough understanding of the various disclosed embodiments. However, those skilled in the art will recognize that embodiments may be practiced without one or more of these specific details. In other instances, well-known apparatuses, structures, and techniques associated with this application may not have been shown or described in detail to avoid unnecessarily obscuring the description of the embodiments.

[0021] Unless the context requires otherwise, throughout the specification and claims, the word “comprising” and its variations, such as “including” and “having”, shall be understood to have an open, inclusive meaning, that is, to be interpreted as “including, but not limited to”.

[0022] Throughout this specification, references to "an embodiment" or "an embodiment" indicate that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Therefore, the appearance of "in an embodiment" or "an embodiment" in various places throughout the specification does not necessarily refer to the same embodiment. Furthermore, a particular feature, structure, or characteristic may be combined in any manner in one or more embodiments.

[0023] The singular forms “a” and “the” used in this specification and the appended claims include plural references unless otherwise expressly stated herein. It should be noted that the term “or” is generally used to mean “and / or” unless otherwise expressly stated herein.

[0024] In the following description, in order to clearly demonstrate the structure and working method of the present invention, a number of directional terms will be used. However, terms such as "front", "back", "left", "right", "outside", "inside", "outward", "inward", "up", and "down" should be understood as convenient terms and not as limiting terms.

[0025] Furthermore, terms such as "horizontal," "vertical," and "sag" do not imply that components must be absolutely horizontal or suspended, but rather that they can be slightly tilted. For example, "horizontal" simply means that its direction is more horizontal relative to "vertical," and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0026] In the description of this application, it should also be noted that, unless otherwise expressly specified and limited, the terms "set up," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0027] This invention provides an online method for estimating vehicle wheel center load. The method can be executed by an electronic device, which includes a processor, a memory, an input interface, and an output interface. The input interface receives vehicle sensor signals; the output interface outputs the predicted vehicle wheel center load.

[0028] The overall flow of this method is as follows Figure 1 As shown, it includes the following steps: S1. The vehicle sensor signals are selected using a multi-task Lasso and stability selection model to obtain multiple dominant sensor signals.

[0029] The number of candidate channels accessible on the vehicle bus is enormous (up to 139 in this scheme), including longitudinal / lateral / vertical acceleration, steering wheel angle, yaw rate, GPS vehicle speed, suspension displacement, and wheel-center triaxial acceleration. Directly using all channels would lead to dimensionality expansion and redundancy. This invention employs multi-task Lasso joint stability selection, with the specific optimization expression as follows. The optimization objective is a row sparse regression of the combined outputs of the six wheel-center loads: in, For candidate sensor signal matrices, This is a six-component wheel center load reference value matrix. The coefficient matrix, This is a row group sparse regularization term that forces the coefficients of the entire row in irrelevant channels to zero simultaneously. This represents the number of training samples; This is the intercept vector of the six-component wheel center load output. for A column vector of all 1s. This indicates that the intercept term is broadcast across all samples; is the regularization coefficient for multi-task Lasso, used to control the tradeoff between fitting error and row group sparsity; The Frobenius norm is used to measure the energy of matrix residuals. For row sparse norm, where Indicates the first Each candidate sensor channel corresponds to a row of output coefficients. The frequency of each channel being selected is calculated by repeatedly performing random subsampling on the data and resolving the problem. ,Will The channels with typical thresholds of 0.6 to 0.8 are used as the final input subset. In this scheme, 8 dominant sensor signals are ultimately retained from 139 candidate signals: longitudinal acceleration, steering wheel angle, yaw rate, GPS vehicle speed, left front suspension displacement, and left front wheel X / Y / Z axis acceleration. This covers longitudinal dynamics, steering and yaw states, suspension kinematics, and wheel-center three-axis acceleration, while avoiding redundancy.

[0030] S2. Construct a non-causal MIMO-FIR model.

[0031] Under any given operating condition, the vehicle dynamics can be approximated as a linear time-invariant system, which can be divided into pseudo-inputs. (The 8 dominant sensor signals selected above) and pseudo-output (6-component wheel center load, The system state-space model is as follows: Among them, road surface excitation Compared with the initial state All are unknown. In the formula, This is the vehicle's equivalent state vector, which can include the displacement, velocity, or acceleration of the vehicle body, suspension, and wheels. It is the differential of the state vector; For unknown road surface excitation input; Here is the state transition matrix. The road surface excitation input matrix; , Pseudo-input The output matrix and the direct transfer matrix; , These are pseudo-outputs. The output matrix and direct transfer matrix are used to illustrate the source of the transfer rate operator, which can be identified from calibration data or obtained through local linearization. Real-time measurement is not required in the online estimation stage. Defined by the transitivity operator ( ),in As a time derivative, this operator establishes a pure output-output relationship between pseudo-inputs and pseudo-outputs, making the estimation independent of road surface measurements. By performing Laurent series truncation, we obtain a non-causal MIMO-FIR model: in , The orders of causality and non-causality are respectively represented. This structure simultaneously captures the influence of pseudo-inputs on wheel center loads at current, past, and future finite moments, enabling it to describe time delays and feedforward responses commonly found in vehicle dynamics. All samples are stacked into a matrix form. , where the parameter matrix , , This is a pseudo-input dimension. For the first The load component of each wheel center in order The corresponding FIR coefficient vector at that point, Indicates non-causal preceding terms. Indicates the current item. Indicates the lagged term of causality; For the first Modeling residuals or measurement noise for each output component.

[0032] Matrix form middle, express The six-component wheel center load matrix corresponding to each time sample; For pseudo input window The regression matrix formed by splicing; The parameter matrix is ​​obtained by stacking all FIR coefficients according to their order and input channels; The residual matrix; , This represents the number of pseudo-input channels.

[0033] To continuously update MIMO-FIR parameters in streaming data without storing the entire history, this invention designs a recursive least squares algorithm based on the Woodbury identity. Given an initial batch dataset... To obtain the initial parameter estimates For the first Batch new data blocks ,definition and order Then the recursive update formula is: This algorithm only needs to invert small matrices with a dimension equal to the batch size, avoiding inversion of large matrices. Direct inversion significantly reduces computational complexity, and all six outputs share the same covariance matrix. This allows the six-component load to be updated synchronously, maintaining the consistency of coupling between the components.

[0034] S3. Extracting working condition features based on principal component analysis (PCA), short-time Fourier transform (STFT), and non-negative matrix factorization (NMF).

[0035] To characterize the non-stationary nature of vehicle dynamic response under different operating conditions, this invention constructs the following operating condition feature (time-frequency feature) extraction process, such as... Figure 2 As shown, the time-frequency feature extraction process for the operating condition includes: S31: Principal component fusion. For the selected... Principal component analysis (PCA) is performed on the road sensor data, and the principal direction corresponding to the largest eigenvalue is selected. 3D signal projection is a single-channel fused signal This compresses dimensionality while preserving maximum variance information. This refers to the single-channel characteristic signal after PCA fusion; Let be the direction vector of the first principal component, satisfying ; for 3D pseudo-input vector; The mean of the pseudo-input vectors within the training window is used to... Centralization is performed. This fusion step is only used for constructing load characteristics and does not change the input definition of the subsequent MIMO-FIR load estimation model.

[0036] S32: Short-Time Fourier Transform (STFT). For STFT is calculated by applying a Hamming window and sliding the calculation to obtain the time-frequency matrix. , This represents the number of frequency grids. The time frame number captures the evolution of the spectrum over time.

[0037] S33: Non-negative Matrix Factorization (NMF). For perform decomposition with non-negative constraints , where is the spectral pattern basis matrix, is the temporal activation coefficient matrix. The number of components is determined by the comprehensive objective of reconstruction error and stability: In the formula, is the candidate number of NMF components, is the final selected number of components; is the amplitude or power time-frequency matrix obtained by STFT; is the Frobenius norm; is the penalty term for the instability of decomposition results under multiple random initializations or sub-samplings; is the stability weight for NMF model selection.

[0038] S34: Statistical feature extraction. For each pair of extract 11 statistical features to form an extreme value vector , with a total of dimensions. The 11 features are respectively: spectral / temporal mean ( ), variance ( ), skewness ( ), which are used to characterize centrality, dispersion and asymmetry respectively; spectral / temporal sparsity ( ), which is used to characterize energy concentration; spectral / temporal discontinuity ( ), which is used to characterize adjacent frame jumps; and energy proportion ( ), which is used to characterize the relative contribution of components. These features comprehensively reflect the spectral center, energy distribution, sparse structure and temporal non-smoothness of working conditions, and have good discrimination ability for changes in road surface roughness, vehicle speed, and maneuver type.

[0039] S4. Identify working conditions based on Extreme Value Machine (EVM).

[0040] For each known working condition , collect its extreme value vector set , and select a representative subset according to a greedy strategy . For each representative extreme value vector , use the empirical distribution of distances between it and other samples of the same working condition to fit a Weibull distribution, and obtain the scale and shape . Define the membership probability (CAP) that the new sample belongs to the working condition : in Use Euclidean or cosine distance. Figure 3 The signal probability distribution based on EVM is shown under different operating conditions.

[0041] When running online, extracting data at the current moment Calculate the inclusion probability for all known operating conditions and make a decision using the following formula: in This is the rejection threshold (typically 0.7 to 0.8). When a known operating condition is identified... When the condition is "new", only the RLS update of the MIMO-FIR model corresponding to that condition is triggered; when the condition is determined to be "new", the system temporarily uses the same data as the current model. The most recently known work condition model is used as a fallback option, while a new work condition data buffer and new model initialization are initiated simultaneously. To prevent the unlimited expansion of invalid new work conditions, the maximum number of concurrently maintained "new work condition" tags is three; if this number is exceeded, the category with the greatest dissimilarity to the known database is prioritized. Coverage threshold. Determined through cross-validation (typically) A balance is achieved between compactness and recognition accuracy.

[0042] S5. Based on the working condition identification results, update the parameters of the non-causal MIMO-FIR model and output the predicted vehicle wheel center load.

[0043] Within each sampling period, the six-component wheel center load estimate is first output. Then, based on the EVM decision, it is determined whether to perform RLS incremental update on the current working condition model. The six-component wheel center load estimate is then output. If it is determined to be a new working condition, the fallback model output is maintained, and the new working condition window is written to the local circular buffer.

[0044] The present invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, wherein the computer program, when executed by the processor, implements the steps of the data processing method described above.

[0045] Electronic devices can be desktop computers, portable computers (laptops), tablet computers (iPads), smart terminals, servers, cloud servers, or industrial control computers, etc.

[0046] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to perform the method of the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0047] The electronic device may include, but is not limited to, a processor and a memory, and may also include input / output devices, network access devices, buses, etc. Specifically, the electronic device may have a data input interface and a data output interface, wherein the data input interface can receive data from the outside via wired or wireless means, and this data may be collected and stored by data acquisition devices.

[0048] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting various parts of the electronic device through various interfaces and lines.

[0049] The memory can be used to store computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory, and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function, etc.; the data storage area may store the aforementioned various data, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0050] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method described above.

[0051] The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium can include any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0052] Compared with the prior art, the present invention has the following advantages: Starting from the causal chain of "unknown road surface excitation → ill-conditioned inverse problem," this invention establishes a pure output-output mapping using a transitivity operator. This eliminates the dependence of wheel center load estimation on road surface profile measurements, fundamentally removing the existing observer's reliance on road surface priors. Therefore, it maintains a stable mapping on any unknown road surface, significantly improving cross-road surface generalization capability. Furthermore, the non-causal FIR structure simultaneously incorporates pseudo-inputs from finite past and future moments, enabling the model to precisely capture time delays and feedforward responses in vehicle dynamics. Compared to structures containing only causal terms, it performs better under transient conditions.

[0053] To address the problem of "closed set assumption → failure under new operating conditions," this invention introduces EVM to construct open set gating and uses Weibull distribution to fit the compact inclusion probability boundary of each class, enabling the system to reject "unseen operating conditions." The causal consequence is that the model no longer makes confident but incorrect estimates of unfamiliar inputs, but instead identifies them as new operating conditions and triggers the initialization and learning of the new model, thereby avoiding the catastrophic generalization failure that is common in existing static models and providing a reliable confidence signal for the upper-level control system.

[0054] To address the problem of "blind or unconditional online updates leading to degradation of steady-state accuracy," the condition-triggered update strategy of this invention updates the corresponding model only when a known operating condition is identified. This avoids noise injection under known operating conditions and also avoids forced correction of erroneous models under new operating conditions. The causal consequence is that the estimation is more stable under steady-state operating conditions and the response is faster under transient new operating conditions, achieving refined adaptation of "updating when it should be updated and maintaining when it should be maintained."

[0055] To address the issue of "channel redundancy → dimensionality and computational pressure," this invention obtains a compact subset of inputs shared across outputs through multi-task Lasso stability selection. The causal consequences are: significant compression of model parameter dimensions, improved identifiability, stable variable selection across multiple trials, and direct reduction of inference latency on edge devices. This allows the entire processing pipeline to be completed in an average of 4.8ms / cycle, meeting the 10ms automotive-grade real-time budget.

[0056] To address the issue of "offline training → inability to adapt to configuration changes," the RLS incremental update of this invention, combined with the Woodbury identity, enables the system to continuously update online without storing all historical data. When the load, tires, or suspension change, conditional gating automatically triggers the update, and the system can recover to near-baseline estimation accuracy within 5 to 15 seconds, providing out-of-the-box full lifecycle adaptive capability.

[0057] Finally, the edge-cloud collaborative architecture enables new operating condition models learned on prototype vehicles to be directly pushed to a fleet of vehicles of the same type. The causal consequence is that group-level learning accelerates model maturity, and individual vehicles do not need to learn all operating conditions independently to enjoy the experience "from other vehicles", which significantly shortens the system deployment and optimization cycle.

[0058] In summary, this invention achieves a systematic improvement over existing technologies in five dimensions: generalization, adaptability, real-time performance, interpretability, and engineering deployability, providing a robust and economical soft measurement solution for chassis perception in next-generation intelligent vehicles.

[0059] The preferred embodiments of the present invention have been described in detail above. However, it should be understood that after reading the above teachings, those skilled in the art can make various alterations or modifications to the present invention. These equivalent forms also fall within the scope defined by the appended claims.

Claims

1. A method for online estimation of vehicle wheel center load, characterized in that, Includes the following steps: S1. The vehicle sensor signals are selected using a multi-task Lasso and stability selection model to obtain multiple dominant sensor signals. S2. Construct a non-causal MIMO-FIR model; S3. Extracting working condition features based on principal component analysis (PCA), short-time Fourier transform (STFT), and nonnegative matrix factorization (NMF); S4. Working conditions are identified based on the extreme value machine EVM; S5. Based on the working condition identification results, update the parameters of the non-causal MIMO-FIR model and output the predicted vehicle wheel center load.

2. The online estimation method for vehicle wheel center load according to claim 1, characterized in that, S1 includes: An optimization objective function is established based on a multi-task Lasso and stability selection model: ， in, For candidate sensor signal matrix, This is a six-component wheel center load reference value matrix. The coefficient matrix, For row group sparse regularization, The number of training samples, This is the intercept vector of the six-component wheel center load output. for A column vector of all 1s. This indicates that the intercept term is broadcast across all samples. represents the regularization coefficient for multi-task Lasso. It is the Frobenius norm. For the row sparse norm, where, Indicates the first Each candidate sensor channel corresponds to a row of output coefficients; By performing multiple random subsamplings on the data and resolving the problem, the frequency of each channel being selected is statistically analyzed. ,Will The channels are used as the final input subset, where, This is a preset threshold.

3. The online estimation method for vehicle wheel center load according to claim 2, characterized in that, It ranges from 0.6 to 0.

8.

4. The online estimation method for vehicle wheel center load according to claim 2, characterized in that, S2 include: Under any given operating condition, the vehicle dynamics can be approximated as a linear time-invariant system, and the system state-space model is as follows: ， ， ， in, This is the vehicle's equivalent state vector. It is the differential of the state vector; For unknown road surface excitation input; Here is the state transition matrix. The road surface excitation input matrix; , Pseudo-input The output matrix and the direct transfer matrix; , These are pseudo-outputs. The output matrix and the direct transfer matrix; Defined by the transitivity operator ,in, As a time derivative, this operator establishes a pure output-output relationship between pseudo-inputs and pseudo-outputs, making the estimation no longer dependent on road surface measurements; right By performing Laurent series truncation, we obtain a non-causal MIMO-FIR model: ， in , They are the causal and non-causal orders, respectively. For the first The load component of each wheel center in order The corresponding FIR coefficient vector at that point, Indicates non-causal preceding terms. Indicates the current item. Indicates the lagged term of causality; For the first Modeling residuals or measurement noise for each output component; Stack all samples into a matrix form. ,in, express The six-component wheel center load matrix corresponding to each time sample; For pseudo input window The regression matrix formed by splicing; The parameter matrix is ​​obtained by stacking all FIR coefficients according to their order and input channels; The residual matrix; , This represents the number of pseudo-input channels.

5. The online estimation method for vehicle wheel center load according to claim 4, characterized in that, S2 also includes: continuously updating the MIMO-FIR model parameters using the recursive least squares (RSL) method. Given an initial batch dataset To obtain the initial parameter estimates ; For the Batch new data blocks ,definition and order Then the recursive update formula is: ， 。 6. The online estimation method for vehicle wheel center load according to claim 5, characterized in that, S3 include: Principal component fusion, for the selected Principal component analysis (PCA) is performed on the road sensor data, and the principal direction corresponding to the largest eigenvalue is selected. 3D signal projection is a single-channel fused signal While preserving the maximum variance information, the dimensionality is compressed, where, This refers to the single-channel characteristic signal after PCA fusion; Let be the direction vector of the first principal component, satisfying ; for 3D pseudo-input vector; The mean of the pseudo-input vectors within the training window is used to... Centralization; Short-Time Fourier Transform (STFT) for STFT is calculated by applying a Hamming window and sliding the calculation to obtain the time-frequency matrix. ,in, This represents the number of frequency grids. For time frames; Nonnegative matrix factorization (NMF) for Decomposition with nonnegativity constraints ,in For the spectral mode basis matrix, The time activation coefficient matrix has the following number of components. Determined by combining reconstruction error and stability objectives: ， in, The number of NMF candidate components, The final number of components; It is the Frobenius norm; This is a penalty term for the instability of the decomposition results under multiple random initializations or subsampling. NMF Stability weights selected for the NMF model; Statistical feature extraction, for each pair Eleven statistical features were extracted to form an extreme value vector. ,common Dimension, among which, , For spectral / time domain mean, For spectral / temporal variance, , For spectral / temporal skewness; For spectral / temporal sparsity; , For spectral / temporal discontinuity; Energy percentage.

7. The online estimation method for vehicle wheel center load according to claim 6, characterized in that, S4 include: For each known working condition Collect its extreme value vector set And select a representative subset according to a greedy strategy. ; For each vector representing an extreme value By utilizing the empirical distribution of its distance from other samples under the same working conditions, a Weibull distribution is fitted to obtain the scale. and shape ; Define new sample This is a working condition. The probability of inclusion: ， ， in, Use Euclidean or cosine distance; When running online, extracting data at the current moment Calculate the inclusion probability for all known operating conditions and make a decision using the following formula: in, The rejection threshold is set when the condition is determined to be a known operating condition. When the condition is "new", only the RLS update of the MIMO-FIR model corresponding to that condition is triggered; when the condition is determined to be "new", the system temporarily uses the same data as the current model. The most recent known operating condition model is used as a fallback option, while a new operating condition data buffer and a new model initialization are started.

8. The online estimation method for vehicle wheel center load according to claim 7, characterized in that, It is 0.7~0.

8.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 8.

Images