A vehicle adaptive steering method and system

Abstract

The application provides a vehicle adaptive steering method and system, which comprises the following steps: collecting historical driving big data under a plurality of vehicle types and a plurality of driving scenes, extracting common steering behavior characteristics and individual driving preference characteristics under different driving scenes through a transfer learning network, and synchronously generating corresponding driving behavior characteristic sets; collecting the characteristics of goods, transportation constraints and safety level requirements of a current transportation task, quantifying a task demand vector by using an entropy weight method, and synchronously generating a corresponding characteristic adaptation matrix based on the driving behavior characteristic sets and the task demand vector; collecting driving data and goods state data of the vehicle through a vehicle-mounted edge computing node, and synchronously generating corresponding target steering parameters in combination with the characteristic adaptation matrix; generating corresponding steering control instructions according to the target steering parameters, and synchronously issuing the steering control instructions to a steering system to complete corresponding steering control. The application can effectively improve the steering efficiency of the vehicle.

Description

Technical Field

[0001] This invention relates to the field of automotive technology, and in particular to a vehicle adaptive steering method and system. Background Technology

[0002] The steering system is a core component ensuring vehicle handling performance and driving safety, and its adaptability directly affects the driving experience and operational safety. Commercial vehicles are widely used in logistics, transportation, and engineering operations, with complex and diverse operating scenarios, including long-distance transportation, urban delivery, and hazardous materials transportation. They are characterized by their large size, wide load fluctuations, and variable center of gravity, placing higher demands on the dynamic adaptability and control precision of the steering system compared to passenger vehicles. Steering requirements vary significantly across different scenarios; for example, long-distance transportation requires a balance between stability and fatigue resistance, urban delivery needs improved low-speed maneuverability, and hazardous materials transportation requires precise and smooth steering.

[0003] Currently, most mainstream traditional steering systems in commercial vehicles (such as mechanical hydraulic power steering and traditional electric power steering) use fixed parameter control, relying solely on preset parameters such as vehicle speed and steering angle, and cannot be dynamically adjusted according to actual working conditions. On the one hand, different drivers have significant differences in steering effort, response speed preferences, and other styles, making it difficult for fixed parameters to match individual needs, which can easily lead to driver fatigue or even misjudgment. On the other hand, commercial vehicle carrying tasks fluctuate greatly, with significant differences in vehicle steering load, center of gravity distribution, and safety requirements under conditions such as empty load, full load, and carrying special cargo. Fixed parameters cannot specifically optimize steering characteristics, making it difficult to ensure cargo safety and potentially increasing energy consumption.

[0004] Furthermore, the development of intelligent and connected vehicle technologies has made it possible to collect driving big data (including driver operation, vehicle status, road conditions, etc.), providing data support for adaptive steering control. However, current technologies have not yet formed a steering control scheme based on the coordinated driving of driving big data and transportation tasks, and cannot achieve multi-objective coordination of driver habit adaptation, cargo safety assurance, and energy consumption optimization. The adaptability defects of traditional steering systems have become a bottleneck restricting the improvement of vehicle performance. Therefore, developing a dynamically adjustable adaptive steering method to solve the shortcomings of existing technologies has become an urgent need in the field of automotive steering, and has significant technological and industrial value. Summary of the Invention

[0005] Therefore, the purpose of this invention is to provide a vehicle adaptive steering method and system to solve the problem that the existing technology cannot coordinate the control of the vehicle's steering system based on driving big data and transportation tasks, which leads to a reduction in vehicle performance.

[0006] The first aspect of the present invention proposes:

[0007] A vehicle adaptive steering method, wherein the method includes:

[0008] Collect historical driving big data from several vehicle models and several driving scenarios, extract common steering behavior features and personalized driving preference features under different driving scenarios through transfer learning networks, and simultaneously generate corresponding driving behavior feature sets;

[0009] Collect the cargo characteristics, transportation constraints and safety level requirements of the current transportation task, quantify the task requirement vector using the entropy weight method, and simultaneously generate the corresponding feature adaptation matrix based on the driving behavior feature set and the task requirement vector.

[0010] The vehicle's driving data and cargo status data are collected by the vehicle-mounted edge computing node, and the corresponding target steering parameters are generated simultaneously by combining the feature adaptation matrix.

[0011] Based on the target steering parameters, a corresponding steering control command is generated, and the steering control command is simultaneously sent to the steering system to complete the corresponding steering control.

[0012] The beneficial effects of this invention are as follows: This technical solution accurately extracts common steering features and personalized driving preferences in different scenarios through transfer learning networks, scientifically quantifies the cargo characteristics, transportation constraints, and safety level requirements of the transportation task using the entropy weight method, and then relies on on-board edge computing nodes to collect data in real time and match feature adaptation matrices to achieve deep collaborative steering control between driving big data and transportation tasks. This effectively overcomes the shortcomings of existing technologies, significantly improves the targeting and real-time performance of steering control, ensures driving safety and stability in different transportation scenarios, reduces cargo damage, optimizes the driving experience, and comprehensively improves vehicle performance and transportation adaptability.

[0013] Furthermore, the step of collecting the cargo characteristics, transportation constraints, and safety level requirements of the current transportation task, and quantifying them using the entropy weight method to obtain the task requirement vector includes:

[0014] The coupling relationship between the cargo characteristics, transportation constraints and safety level requirements is detected by the association rule mining algorithm, and composite constraint features are extracted based on the coupling relationship to create a corresponding task requirement index set.

[0015] The initial information entropy weights corresponding to each indicator in the task requirement indicator set are calculated by the entropy weight method, and a dynamic adaptation factor for the transportation scenario is introduced to construct a mapping model between the adaptation factor and the indicator weights.

[0016] The task requirement index set is dynamically modified using the mapping model to generate the corresponding task requirement vector.

[0017] Furthermore, the step of dynamically correcting the task requirement index set through the mapping model to generate the corresponding task requirement vector includes:

[0018] The correction parameters of each indicator output by the mapping model are used as evidence. A credibility allocation function is constructed by combining the fluctuation of cargo characteristics and the range of changes in transportation constraints. At the same time, each correction indicator in the task requirement indicator set is quantified through the credibility allocation function to generate the corresponding preliminary correction indicator set.

[0019] The numerical and categorical indicators in the preliminary correction indicator set are transformed into multimodal feature vectors. Attention mechanism is used to calculate the correlation weights between the multimodal feature vectors to perform cross-dimensional optimization and generate a reinforcement correction indicator set.

[0020] Using the enhanced correction index set as input and the historical best fit vector as a reference, the task requirement vector is trained accordingly through adversarial training of a generative adversarial network.

[0021] Furthermore, the step of generating a corresponding feature adaptation matrix based on the driving behavior feature set and the task requirement vector includes:

[0022] The driving behavior feature set and the indicators of each dimension of the task requirement vector are respectively used as graph nodes to construct a heterogeneous feature association graph. Simultaneously, the potential association strength between nodes is learned through a graph neural network to generate the corresponding graph feature matrix based on the output association representation.

[0023] Using the graph feature matrix as input, a sparse Bayesian learning model is constructed, with the goal of minimizing the feature adaptation error, and the corresponding corrected feature matrix is ​​output.

[0024] By introducing task requirement priority constraints, the modified feature matrix is ​​iteratively solved to generate the corresponding feature adaptation matrix.

[0025] Furthermore, the step of introducing task requirement priority constraints and iteratively solving the modified feature matrix to generate the corresponding feature adaptation matrix includes:

[0026] The Analytic Hierarchy Process (AHP) is introduced to construct a priority judgment matrix for task requirements. Simultaneously, priority quantification coefficients for each requirement dimension are calculated based on the priority judgment matrix, and the priority quantification coefficients are integrated into a corresponding priority constraint matrix.

[0027] Using the product of the modified feature matrix and the priority constraint matrix as the initial input, an iterative solution framework based on the alternating direction multiplier method is constructed. In each iteration, Kalman filtering is introduced to dynamically estimate the solution bias in order to obtain the preliminary feature fitting matrix.

[0028] Multiple sets of transportation scenario disturbance samples are generated using Monte Carlo simulation. The preliminary feature adaptation matrix is ​​then substituted into each sample for adaptation verification. Once the verification is successful, the corresponding feature adaptation matrix is ​​generated.

[0029] Furthermore, the step of collecting vehicle driving data and cargo status data through onboard edge computing nodes and simultaneously combining the feature adaptation matrix to generate corresponding target steering parameters includes:

[0030] The spatiotemporal alignment algorithm eliminates the data misalignment between the driving data and the cargo status data, and the isolated forest algorithm is used to filter abnormal data. The weight coefficients of the two types of data are dynamically allocated according to the characteristics of the cargo, and the corresponding standard data matrix is ​​generated synchronously.

[0031] The standard data matrix is ​​decomposed into a driving state sub-matrix and a cargo state sub-matrix, and the degree of fit between the driving state sub-matrix and the cargo state sub-matrix and the feature fitting matrix is ​​calculated simultaneously using a dual-metric method.

[0032] Based on the fitness level, a corresponding candidate feature vector is generated, and the target steering parameters are generated based on the candidate feature vector.

[0033] Furthermore, the step of generating a corresponding candidate feature vector based on the fit, and then generating the target steering parameters based on the candidate feature vector, includes:

[0034] By combining the specific characteristics of the goods with the transportation constraints, corresponding evaluation factors are set, and the fit degree is dynamically weighted by the fuzzy hierarchical analysis algorithm. At the same time, the core feature dimension is extracted by the kernel principal component analysis algorithm to generate the corresponding candidate feature vector.

[0035] Based on vehicle dynamics rules, intermediate feature vectors that meet preset requirements are selected from the candidate feature vectors.

[0036] Based on a reinforcement learning framework, with the goals of minimizing cargo center of gravity shift and minimizing path tracking error, the optimal steering parameter mapping relationship is obtained by combining actual road condition feedback. Simultaneously, the intermediate feature vector is dynamically compensated and calibrated through the optimal steering parameter mapping relationship to generate the target steering parameters accordingly.

[0037] The second aspect of the present invention proposes:

[0038] An adaptive steering system for a vehicle, wherein the system includes:

[0039] The data acquisition module is used to collect historical driving big data for several vehicle models and several driving scenarios. It extracts common steering behavior features and personalized driving preference features under different driving scenarios through a transfer learning network, and generates corresponding driving behavior feature sets in a synchronous manner.

[0040] The generation module is used to collect the cargo characteristics, transportation constraints and safety level requirements of the current transportation task, quantify the task requirement vector using the entropy weight method, and simultaneously generate the corresponding feature adaptation matrix based on the driving behavior feature set and the task requirement vector.

[0041] The computing module is used to collect vehicle driving data and cargo status data through the vehicle edge computing node, and simultaneously combine the feature adaptation matrix to generate the corresponding target steering parameters.

[0042] The control module is used to generate corresponding steering control commands based on the target steering parameters, and synchronously send the steering control commands to the steering system to complete the corresponding steering control.

[0043] Furthermore, the generation module is specifically used for:

[0044] The coupling relationship between the cargo characteristics, transportation constraints and safety level requirements is detected by the association rule mining algorithm, and composite constraint features are extracted based on the coupling relationship to create a corresponding task requirement index set.

[0045] The initial information entropy weights corresponding to each indicator in the task requirement indicator set are calculated by the entropy weight method, and a dynamic adaptation factor for the transportation scenario is introduced to construct a mapping model between the adaptation factor and the indicator weights.

[0046] The task requirement index set is dynamically modified using the mapping model to generate the corresponding task requirement vector.

[0047] Furthermore, the generation module is specifically used for:

[0048] The correction parameters of each indicator output by the mapping model are used as evidence. A credibility allocation function is constructed by combining the fluctuation of cargo characteristics and the range of changes in transportation constraints. At the same time, each correction indicator in the task requirement indicator set is quantified through the credibility allocation function to generate the corresponding preliminary correction indicator set.

[0049] The numerical and categorical indicators in the preliminary correction indicator set are transformed into multimodal feature vectors. Attention mechanism is used to calculate the correlation weights between the multimodal feature vectors to perform cross-dimensional optimization and generate a reinforcement correction indicator set.

[0050] Using the enhanced correction index set as input and the historical best fit vector as a reference, the task requirement vector is trained accordingly through adversarial training of a generative adversarial network.

[0051] Furthermore, the calculation module is specifically used for:

[0052] The driving behavior feature set and the indicators of each dimension of the task requirement vector are respectively used as graph nodes to construct a heterogeneous feature association graph. Simultaneously, the potential association strength between nodes is learned through a graph neural network to generate the corresponding graph feature matrix based on the output association representation.

[0053] Using the graph feature matrix as input, a sparse Bayesian learning model is constructed, with the goal of minimizing the feature adaptation error, and the corresponding corrected feature matrix is ​​output.

[0054] By introducing task requirement priority constraints, the modified feature matrix is ​​iteratively solved to generate the corresponding feature adaptation matrix.

[0055] Furthermore, the calculation module is specifically used for:

[0056] The Analytic Hierarchy Process (AHP) is introduced to construct a priority judgment matrix for task requirements. Simultaneously, priority quantification coefficients for each requirement dimension are calculated based on the priority judgment matrix, and the priority quantification coefficients are integrated into a corresponding priority constraint matrix.

[0057] Using the product of the modified feature matrix and the priority constraint matrix as the initial input, an iterative solution framework based on the alternating direction multiplier method is constructed. In each iteration, Kalman filtering is introduced to dynamically estimate the solution bias in order to obtain the preliminary feature fitting matrix.

[0058] Multiple sets of transportation scenario disturbance samples are generated using Monte Carlo simulation. The preliminary feature adaptation matrix is ​​then substituted into each sample for adaptation verification. Once the verification is successful, the corresponding feature adaptation matrix is ​​generated.

[0059] Furthermore, the calculation module is specifically used for:

[0060] The spatiotemporal alignment algorithm eliminates the data misalignment between the driving data and the cargo status data, and the isolated forest algorithm is used to filter abnormal data. The weight coefficients of the two types of data are dynamically allocated according to the characteristics of the cargo, and the corresponding standard data matrix is ​​generated synchronously.

[0061] The standard data matrix is ​​decomposed into a driving state sub-matrix and a cargo state sub-matrix, and the degree of fit between the driving state sub-matrix and the cargo state sub-matrix and the feature fitting matrix is ​​calculated simultaneously using a dual-metric method.

[0062] Based on the fitness level, a corresponding candidate feature vector is generated, and the target steering parameters are generated based on the candidate feature vector.

[0063] Furthermore, the calculation module is specifically used for:

[0064] By combining the specific characteristics of the goods with the transportation constraints, corresponding evaluation factors are set, and the fit degree is dynamically weighted by the fuzzy hierarchical analysis algorithm. At the same time, the core feature dimension is extracted by the kernel principal component analysis algorithm to generate the corresponding candidate feature vector.

[0065] Based on vehicle dynamics rules, intermediate feature vectors that meet preset requirements are selected from the candidate feature vectors.

[0066] Based on a reinforcement learning framework, with the goals of minimizing cargo center of gravity shift and minimizing path tracking error, the optimal steering parameter mapping relationship is obtained by combining actual road condition feedback. Simultaneously, the intermediate feature vector is dynamically compensated and calibrated through the optimal steering parameter mapping relationship to generate the target steering parameters accordingly.

[0067] The third aspect of the present invention proposes:

[0068] A computer includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the vehicle adaptive steering method as described above.

[0069] The fourth aspect of the present invention proposes:

[0070] A readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the vehicle adaptive steering method as described above.

[0071] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0072] Figure 1 A flowchart of the vehicle adaptive steering method provided in the first embodiment of the present invention;

[0073] Figure 2 This is a structural block diagram of a vehicle adaptive steering system provided in the third embodiment of the present invention.

[0074] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0075] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0076] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0077] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0078] Please see Figure 1 The figure shows a vehicle adaptive steering method provided in the first embodiment of the present invention. The vehicle adaptive steering method provided in this embodiment can accurately control the vehicle's steering system by combining driving data with actual transportation tasks, thereby improving steering efficiency.

[0079] Specifically, this embodiment provides:

[0080] A vehicle adaptive steering method, wherein the method includes:

[0081] Step S10: Collect historical driving big data for several vehicle models and several driving scenarios, extract common steering behavior features and personalized driving preference features under different driving scenarios through transfer learning network, and generate corresponding driving behavior feature sets simultaneously.

[0082] It should be noted that, firstly, to address the significant differences in driving behavior across different vehicle types (such as light trucks and heavy trucks) and driving scenarios (such as highways, mountain roads, and urban roads), historical driving big data (such as steering angle, vehicle speed, braking frequency, and steering wheel operation force) from multiple vehicle types and scenarios is collected. A transfer learning network is then used to extract common steering behavior features (such as the common pattern of "smaller steering angle at high speeds" across all scenarios) and personalized driving preference features (such as the individual difference of a driver having "more agile steering response on mountain roads"), generating a driving behavior feature set. Specifically, transfer learning can effectively utilize shared features from different vehicle types / scenarios, solving the problem of insufficient data for a single vehicle type / scenario. This feature set provides a "driving behavior benchmark" for subsequent steering parameter adaptation.

[0083] Step S20: Collect the cargo characteristics, transportation constraints and safety level requirements of the current transportation task, quantify the task requirement vector using the entropy weight method, and simultaneously generate the corresponding feature adaptation matrix based on the driving behavior feature set and the task requirement vector.

[0084] It should be noted that, secondly, considering that cargo characteristics (such as fragile goods, heavy goods, and bulk cargo), transportation constraints (such as timeliness requirements and route restrictions), and safety level requirements (such as high safety levels for dangerous goods transportation) directly affect turning strategies (e.g., fragile goods transportation requires gentle turning, and heavy goods transportation requires suppression of lateral tilting), the entropy weight method is used to objectively quantify these qualitative and quantitative requirements, generating a task requirement vector. Specifically, the entropy weight method can allocate weights according to the dispersion of indicators, avoiding the bias of subjective weighting. The task requirement vector transforms abstract transportation requirements into calculable numerical indicators, realizing the connection between "transportation requirements and turning parameters".

[0085] Step S30: Collect vehicle driving data and cargo status data through the vehicle edge computing node, and simultaneously generate corresponding target steering parameters by combining the feature adaptation matrix.

[0086] It should be noted that, next, vehicle driving data (such as real-time vehicle speed, steering angle, and lateral acceleration) and cargo status data (such as cargo center of gravity offset and fixed status monitoring data) are collected in real time through on-board edge computing nodes. Specifically, edge computing has the advantages of low latency and high real-time performance, which can meet the data processing speed requirements of on-board scenarios. Combined with the feature adaptation matrix built in the early stage (characterizing the adaptation relationship between driving behavior features and task requirements), the real-time data is matched and calculated with the adaptation matrix to generate target steering parameters (such as steering ratio coefficient, damping coefficient, and response speed) that fit the current working conditions. Specifically, the introduction of real-time data ensures the dynamic adjustment of steering parameters and avoids the defect of fixed parameters being unable to adapt to changes in working conditions.

[0087] Step S40: Generate a corresponding steering control command based on the target steering parameters, and simultaneously send the steering control command to the steering system to complete the corresponding steering control.

[0088] It should be noted that, finally, the target steering parameters are converted into standardized steering control commands (such as current control commands for electric power steering systems and angle control commands for steer-by-wire systems), and sent to the vehicle steering system (such as EPS and SBW) to drive the steering actuator to complete the steering action. The entire process forms a closed loop of "historical feature foundation - real-time data drive - parameter dynamic optimization - precise control execution", realizing the adaptive response of the steering system to multi-dimensional factors.

[0089] Second Embodiment

[0090] Furthermore, the step of collecting the cargo characteristics, transportation constraints, and safety level requirements of the current transportation task, and quantifying them using the entropy weight method to obtain the task requirement vector includes:

[0091] The coupling relationship between the cargo characteristics, transportation constraints and safety level requirements is detected by the association rule mining algorithm, and composite constraint features are extracted based on the coupling relationship to create a corresponding task requirement index set.

[0092] The initial information entropy weights corresponding to each indicator in the task requirement indicator set are calculated by the entropy weight method, and a dynamic adaptation factor for the transportation scenario is introduced to construct a mapping model between the adaptation factor and the indicator weights.

[0093] The task requirement index set is dynamically modified using the mapping model to generate the corresponding task requirement vector.

[0094] It should be noted that, firstly, cargo characteristics, transportation constraints, and safety level requirements do not exist in isolation, but rather are strongly coupled (for example, the high safety level requirement for "dangerous goods transportation" constrains the transportation constraint of "transportation speed," while simultaneously requiring an upgrade in the "fixed stability" indicator of cargo characteristics). Therefore, association rule mining algorithms (such as the Apriori algorithm) are used to uncover the hidden coupling relationships among the three, extracting composite constraint features (such as "high safety level + fragile goods → turning smoothness ≥90%, turning response delay ≤0.2s"), and creating a set of task requirement indicators. Specifically, the extraction of composite constraint features avoids the omission of requirements caused by the independent quantification of single indicators, ensuring the comprehensiveness of the indicator set.

[0095] Secondly, the entropy weight method is used to calculate the initial information entropy weight of each indicator in the task requirement indicator set. The entropy weight method determines the weight by analyzing the dispersion of indicator values. For example, the "cargo center of gravity height" indicator has a large dispersion (affecting the risk of steering roll), so it is given a higher weight to ensure the objectivity of weight allocation. However, the transportation scenario is dynamic (such as the different priorities of transportation constraints for the same cargo on highways and mountain roads). Therefore, dynamic adaptation factors of transportation scenarios (such as road type factors and vehicle speed factors) are introduced to construct a mapping model between adaptation factors and indicator weights. Specifically, this model can dynamically adjust the indicator weights according to the real-time scenario. For example, the weight of the "steering flexibility" indicator is increased in the mountain road scenario, and the weight of the "steering stability" indicator is increased in the highway scenario.

[0096] Finally, the task requirement indicator set is dynamically corrected through a mapping model: for example, in high-speed scenarios, the weight of the "steering smoothness" indicator is adjusted from 0.3 to 0.5, and the weight of the "steering response speed" indicator is adjusted from 0.2 to 0.1 through the mapping model; the corrected indicator weights and quantified values ​​are integrated to generate a task requirement vector. Specifically, the dynamically corrected requirement vector can accurately match real-time transportation scenarios, providing accurate requirement guidance for subsequent feature adaptation.

[0097] Furthermore, the step of dynamically correcting the task requirement index set through the mapping model to generate the corresponding task requirement vector includes:

[0098] The correction parameters of each indicator output by the mapping model are used as evidence. A credibility allocation function is constructed by combining the fluctuation of cargo characteristics and the range of changes in transportation constraints. At the same time, each correction indicator in the task requirement indicator set is quantified through the credibility allocation function to generate the corresponding preliminary correction indicator set.

[0099] The numerical and categorical indicators in the preliminary correction indicator set are transformed into multimodal feature vectors. Attention mechanism is used to calculate the correlation weights between the multimodal feature vectors to perform cross-dimensional optimization and generate a reinforcement correction indicator set.

[0100] Using the enhanced correction index set as input and the historical best fit vector as a reference, the task requirement vector is trained accordingly through adversarial training of a generative adversarial network.

[0101] It should be noted that, firstly, the correction parameters of each indicator output by the mapping model are affected by fluctuations in cargo characteristics (such as deviations in cargo loading) and changes in transportation constraints (such as adjustments to timeliness requirements due to unforeseen road conditions), resulting in uncertainty. Therefore, the correction parameters are used as evidence, and a credibility allocation function (such as the basic probability allocation function in DS evidence theory) is constructed by combining the fluctuation range and the magnitude of change. The credibility of each correction indicator is quantified through this function (e.g., when the cargo loading deviation is ≤5%, the credibility of the correction indicator is 0.9; when the deviation is >10%, the credibility drops to 0.6), generating a preliminary set of correction indicators. Specifically, credibility quantification can reduce the impact of uncertainties on demand indicators and ensure the reliability of the indicator set.

[0102] Secondly, the task requirement indicators include numerical indicators (such as cargo weight and center of gravity height) and categorical indicators (such as cargo type and safety level). The two types of indicators have different feature dimensions, and direct fusion can easily lead to information distortion. Therefore, they are transformed into multimodal feature vectors, and an attention mechanism is used to calculate the correlation weight between each vector (such as the correlation weight between the "fragile" category vector and the "turning smoothness" numerical vector is 0.8). By optimizing and integrating the information of different types of indicators across dimensions, a set of enhanced and corrected indicators is generated. Specifically, the attention mechanism can focus on key correlations, avoid interference from irrelevant dimensions, and improve the information density of the indicator set.

[0103] Finally, using the enhanced correction index set as input and the optimal adaptation vector of "steering parameters - transportation demand" in historical transportation scenarios as a reference, adversarial training is performed through a generative adversarial network (GAN): the generator continuously generates candidate demand vectors, and the discriminator compares the candidate vectors with the historical optimal adaptation vectors. Through iterative adversarial optimization of the distribution of candidate vectors, a task demand vector that is consistent with the historical optimal adaptation pattern and adapts to the current transportation demand is finally trained. Specifically, adversarial training can ensure that the demand vector not only conforms to the general adaptation pattern, but also has the personalized adaptation capability of the current scenario.

[0104] Furthermore, the step of generating a corresponding feature adaptation matrix based on the driving behavior feature set and the task requirement vector includes:

[0105] The driving behavior feature set and the indicators of each dimension of the task requirement vector are respectively used as graph nodes to construct a heterogeneous feature association graph. Simultaneously, the potential association strength between nodes is learned through a graph neural network to generate the corresponding graph feature matrix based on the output association representation.

[0106] Using the graph feature matrix as input, a sparse Bayesian learning model is constructed, with the goal of minimizing the feature adaptation error, and the corresponding corrected feature matrix is ​​output.

[0107] By introducing task requirement priority constraints, the modified feature matrix is ​​iteratively solved to generate the corresponding feature adaptation matrix.

[0108] It should be noted that, firstly, the driving behavior feature set (such as the rate of change of steering angle and the frequency of steering wheel operation) and the task requirement vector (such as cargo stability requirements and steering smoothness requirements) have different indicator dimensions, belonging to heterogeneous features. Direct matching may easily overlook potential correlations. Therefore, the indicators of each dimension of the two types of features are used as graph nodes to construct a heterogeneous feature correlation graph. The weight of the edge in the graph represents the correlation strength between nodes (e.g., the correlation weight between the "steering smoothness requirement" node and the "steering angle change rate feature" node is 0.7). The potential correlation strength between nodes is learned through a graph neural network (GNN), outputting a high-dimensional correlation representation, and then generating a graph feature matrix. Specifically, GNN is good at mining deep correlations of heterogeneous features, and the graph feature matrix can comprehensively characterize the coupling relationship between "driving behavior and task requirements".

[0109] Secondly, the graph feature matrix may contain redundant features and noise interference, affecting the fitting accuracy. Therefore, a sparse Bayesian learning model is constructed using the graph feature matrix as input, with the goal of minimizing the feature fitting error (such as minimizing the matching deviation of "driving behavior features - task requirements"). The model automatically filters key features and removes redundant information, outputting a corrected feature matrix. Specifically, the sparse Bayesian learning model has both feature selection and parameter estimation capabilities, which can improve the sparsity and accuracy of the feature matrix.

[0110] Finally, different transportation tasks have different priorities (e.g., in the transportation of dangerous goods, "safety and stability" has a higher priority than "driving comfort"). Therefore, task requirement priority constraints are introduced, and the modified feature matrix is ​​substituted into the constraint conditions for iterative solution. For example, during the iteration process, the feature matching corresponding to the high-priority requirements is satisfied first, and the element values ​​of the feature matrix are adjusted. Finally, a feature adaptation matrix is ​​generated. Specifically, the introduction of priority constraints ensures that the adaptation matrix responds to the core requirements first, and improves the adaptation of steering parameters.

[0111] Furthermore, the step of introducing task requirement priority constraints and iteratively solving the modified feature matrix to generate the corresponding feature adaptation matrix includes:

[0112] The Analytic Hierarchy Process (AHP) is introduced to construct a priority judgment matrix for task requirements. Simultaneously, priority quantification coefficients for each requirement dimension are calculated based on the priority judgment matrix, and the priority quantification coefficients are integrated into a corresponding priority constraint matrix.

[0113] Using the product of the modified feature matrix and the priority constraint matrix as the initial input, an iterative solution framework based on the alternating direction multiplier method is constructed. In each iteration, Kalman filtering is introduced to dynamically estimate the solution bias in order to obtain the preliminary feature fitting matrix.

[0114] Multiple sets of transportation scenario disturbance samples are generated using Monte Carlo simulation. The preliminary feature adaptation matrix is ​​then substituted into each sample for adaptation verification. Once the verification is successful, the corresponding feature adaptation matrix is ​​generated.

[0115] It should be noted that, firstly, the Analytic Hierarchy Process (AHP) is used to construct a priority judgment matrix for task requirements: the task requirements are divided into a target layer (such as "transportation safety", "driving efficiency", "cargo protection") and a criterion layer (such as "steering stability" and "steering response speed"). The relative importance of each requirement dimension is determined by pairwise comparisons, and priority quantification coefficients are calculated (such as the priority coefficient of "cargo protection" being 0.6 and "driving efficiency" being 0.3). This is then integrated into a priority constraint matrix. Specifically, AHP can transform subjective priority judgments into objective quantification coefficients to ensure the rationality of the constraint matrix.

[0116] Secondly, using the product of the modified feature matrix and the priority constraint matrix as the initial input, an iterative solution framework based on the Alternating Direction Multiplier Method (ADMM) is constructed: the ADMM algorithm can decompose complex constrained optimization problems into multiple simple subproblems, improving the solution efficiency; in each iteration, Kalman filtering is introduced to dynamically estimate the solution deviation (such as the real-time estimation of "feature matching deviation" during the iteration process), and the iteration direction of the feature matrix is ​​adjusted according to the deviation to gradually approach the optimal solution and obtain the preliminary feature fitting matrix. Specifically, Kalman filtering has the ability to dynamically estimate deviations, which can improve the stability and accuracy of iterative solutions.

[0117] Finally, the transportation scenario is subject to uncertainties and disturbances (such as sudden road conditions and cargo center of gravity shifts), requiring verification of the robustness of the preliminary feature fitting matrix. Therefore, Monte Carlo simulation is used to generate multiple sets of transportation scenario disturbance samples (such as cargo center of gravity shifts of ±10% and vehicle speed fluctuations of ±20%). The preliminary feature fitting matrix is ​​then substituted into each sample for fit verification (such as verifying whether the feature matching error of the matrix under the disturbance scenario is ≤ a preset threshold). If the verification results of all samples meet the requirements, the final feature fitting matrix is ​​output. Specifically, Monte Carlo simulation can comprehensively cover potential disturbance scenarios, ensuring the robustness of the fitting matrix.

[0118] Furthermore, the step of collecting vehicle driving data and cargo status data through onboard edge computing nodes and simultaneously combining the feature adaptation matrix to generate corresponding target steering parameters includes:

[0119] The spatiotemporal alignment algorithm eliminates the data misalignment between the driving data and the cargo status data, and the isolated forest algorithm is used to filter abnormal data. The weight coefficients of the two types of data are dynamically allocated according to the characteristics of the cargo, and the corresponding standard data matrix is ​​generated synchronously.

[0120] The standard data matrix is ​​decomposed into a driving state sub-matrix and a cargo state sub-matrix, and the degree of fit between the driving state sub-matrix and the cargo state sub-matrix and the feature fitting matrix is ​​calculated simultaneously using a dual-metric method.

[0121] Based on the fitness level, a corresponding candidate feature vector is generated, and the target steering parameters are generated based on the candidate feature vector.

[0122] It should be noted that, firstly, the driving data and cargo status data collected by the vehicle-mounted edge computing node may have time misalignment (e.g., driving data is collected at time t, while cargo status data is collected at t+0.1s) and abnormal data (e.g., jump values ​​caused by sensor malfunctions). Therefore, a spatiotemporal alignment algorithm (e.g., Dynamic Time Warping (DTW)) is used to eliminate data misalignment and ensure the time synchronization of the two types of data. An isolated forest algorithm is used to filter abnormal data and eliminate invalid interference. Simultaneously, the weight coefficients of the two types of data are dynamically allocated according to cargo characteristics (e.g., for heavy cargo transportation, the weight of cargo status data is 0.6, and the weight of driving data is 0.4; for ordinary cargo transportation, the weights are adjusted to 0.4 and 0.6 respectively), generating a standard data matrix. Specifically, spatiotemporal alignment, anomaly filtering, and dynamic weighting ensure the accuracy and relevance of the data matrix.

[0123] Secondly, the standard data matrix is ​​decomposed into driving state sub-matrices (such as vehicle speed, steering angle, and lateral acceleration) and cargo state sub-matrices (such as center of gravity offset and fixation strength) according to data type. A dual-metric method (such as cosine similarity + Euclidean distance) is used to calculate the fit between the two sub-matrices and the feature fitting matrix: cosine similarity measures the consistency of feature directions, and Euclidean distance measures the closeness of feature values. The combination of the two can comprehensively quantify the matching degree between real-time data and the fitting matrix. Specifically, the dual-metric method avoids the one-sidedness of a single metric and improves the accuracy of the fit calculation.

[0124] Finally, candidate feature vectors are generated based on the fit ranking (e.g., the top 5 feature combinations). The candidate feature vectors contain the core information of "real-time driving status - real-time cargo status - feature fit rules". Through the mapping relationship between feature vectors and steering parameters (e.g., steering ratio coefficient = weighted sum of elements in candidate feature vectors), the corresponding target steering parameters are generated. Specifically, the selection of candidate feature vectors can focus on the optimal matching scheme to ensure the fit of steering parameters.

[0125] Furthermore, the step of generating a corresponding candidate feature vector based on the fit, and then generating the target steering parameters based on the candidate feature vector, includes:

[0126] By combining the specific characteristics of the goods with the transportation constraints, corresponding evaluation factors are set, and the fit degree is dynamically weighted by the fuzzy hierarchical analysis algorithm. At the same time, the core feature dimension is extracted by the kernel principal component analysis algorithm to generate the corresponding candidate feature vector.

[0127] Based on vehicle dynamics rules, intermediate feature vectors that meet preset requirements are selected from the candidate feature vectors.

[0128] Based on a reinforcement learning framework, with the goals of minimizing cargo center of gravity shift and minimizing path tracking error, the optimal steering parameter mapping relationship is obtained by combining actual road condition feedback. Simultaneously, the intermediate feature vector is dynamically compensated and calibrated through the optimal steering parameter mapping relationship to generate the target steering parameters accordingly.

[0129] It should be noted that, firstly, evaluation factors (such as shock suppression factors and flexibility factors) are set based on cargo characteristics (e.g., fragile goods require shock suppression) and transportation constraints (e.g., mountain roads require improved flexibility). Fuzzy Hierarchical Analysis (FAHP) is then used to dynamically weight the fit (e.g., when transporting fragile goods, the shock suppression factor has a weight of 0.5, and the weighted fit tends to favor smoother turns). Kernel Principal Component Analysis (KPCA) is then used to extract core feature dimensions, remove redundant information, and generate candidate feature vectors. Specifically, FAHP can handle the fuzziness of evaluation factors, while KPCA excels at handling nonlinear features; combining the two improves the quality of candidate vectors.

[0130] Secondly, vehicle steering parameters are subject to physical constraints of vehicle dynamics (such as steering angle range ±45°, steering speed ≤50° / s). Therefore, by combining vehicle dynamics rules (such as the roll stability model based on Newtonian mechanics), intermediate feature vectors that meet physical constraints and safety requirements are selected from the candidate feature vectors. Specifically, the selection process can prevent steering parameters from exceeding the vehicle performance limits and ensure the feasibility of the parameters.

[0131] Finally, a steering parameter optimization model is constructed based on the reinforcement learning (RL) framework: the reward functions are "minimum cargo center of gravity offset" (ensuring cargo stability) and "minimum path tracking error" (ensuring driving accuracy), and the state input is real-time road condition feedback (such as road surface adhesion coefficient and lateral acceleration feedback). The optimal steering parameter mapping relationship is learned iteratively by the agent. The intermediate feature vector is dynamically compensated and calibrated using this mapping relationship (such as calibrating steering parameters to reduce the risk of sideslip when the road surface adhesion coefficient is low). Finally, the target steering parameters are generated. Specifically, reinforcement learning can realize real-time dynamic optimization of steering parameters, and road condition feedback calibration ensures the adaptive ability of parameters to complex road conditions.

[0132] Please see Figure 2 The third embodiment of the present invention provides:

[0133] An adaptive steering system for a vehicle, wherein the system includes:

[0134] The data acquisition module is used to collect historical driving big data for several vehicle models and several driving scenarios. It extracts common steering behavior features and personalized driving preference features under different driving scenarios through a transfer learning network, and generates corresponding driving behavior feature sets in a synchronous manner.

[0135] The generation module is used to collect the cargo characteristics, transportation constraints and safety level requirements of the current transportation task, quantify the task requirement vector using the entropy weight method, and simultaneously generate the corresponding feature adaptation matrix based on the driving behavior feature set and the task requirement vector.

[0136] The computing module is used to collect vehicle driving data and cargo status data through the vehicle edge computing node, and simultaneously combine the feature adaptation matrix to generate the corresponding target steering parameters.

[0137] The control module is used to generate corresponding steering control commands based on the target steering parameters, and synchronously send the steering control commands to the steering system to complete the corresponding steering control.

[0138] Furthermore, the generation module is specifically used for:

[0139] The coupling relationship between the cargo characteristics, transportation constraints and safety level requirements is detected by the association rule mining algorithm, and composite constraint features are extracted based on the coupling relationship to create a corresponding task requirement index set.

[0140] The initial information entropy weights corresponding to each indicator in the task requirement indicator set are calculated by the entropy weight method, and a dynamic adaptation factor for the transportation scenario is introduced to construct a mapping model between the adaptation factor and the indicator weights.

[0141] The task requirement index set is dynamically modified using the mapping model to generate the corresponding task requirement vector.

[0142] Furthermore, the generation module is specifically used for:

[0143] The correction parameters of each indicator output by the mapping model are used as evidence. A credibility allocation function is constructed by combining the fluctuation of cargo characteristics and the range of changes in transportation constraints. At the same time, each correction indicator in the task requirement indicator set is quantified through the credibility allocation function to generate the corresponding preliminary correction indicator set.

[0144] The numerical and categorical indicators in the preliminary correction indicator set are transformed into multimodal feature vectors. Attention mechanism is used to calculate the correlation weights between the multimodal feature vectors to perform cross-dimensional optimization and generate a reinforcement correction indicator set.

[0145] Using the enhanced correction index set as input and the historical best fit vector as a reference, the task requirement vector is trained accordingly through adversarial training of a generative adversarial network.

[0146] Furthermore, the calculation module is specifically used for:

[0147] The driving behavior feature set and the indicators of each dimension of the task requirement vector are respectively used as graph nodes to construct a heterogeneous feature association graph. Simultaneously, the potential association strength between nodes is learned through a graph neural network to generate the corresponding graph feature matrix based on the output association representation.

[0148] Using the graph feature matrix as input, a sparse Bayesian learning model is constructed, with the goal of minimizing the feature adaptation error, and the corresponding corrected feature matrix is ​​output.

[0149] By introducing task requirement priority constraints, the modified feature matrix is ​​iteratively solved to generate the corresponding feature adaptation matrix.

[0150] Furthermore, the calculation module is specifically used for:

[0151] The Analytic Hierarchy Process (AHP) is introduced to construct a priority judgment matrix for task requirements. Simultaneously, priority quantification coefficients for each requirement dimension are calculated based on the priority judgment matrix, and the priority quantification coefficients are integrated into a corresponding priority constraint matrix.

[0152] Using the product of the modified feature matrix and the priority constraint matrix as the initial input, an iterative solution framework based on the alternating direction multiplier method is constructed. In each iteration, Kalman filtering is introduced to dynamically estimate the solution bias in order to obtain the preliminary feature fitting matrix.

[0153] Multiple sets of transportation scenario disturbance samples are generated using Monte Carlo simulation. The preliminary feature adaptation matrix is ​​then substituted into each sample for adaptation verification. Once the verification is successful, the corresponding feature adaptation matrix is ​​generated.

[0154] Furthermore, the calculation module is specifically used for:

[0155] The spatiotemporal alignment algorithm eliminates the data misalignment between the driving data and the cargo status data, and the isolated forest algorithm is used to filter abnormal data. The weight coefficients of the two types of data are dynamically allocated according to the characteristics of the cargo, and the corresponding standard data matrix is ​​generated synchronously.

[0156] The standard data matrix is ​​decomposed into a driving state sub-matrix and a cargo state sub-matrix, and the degree of fit between the driving state sub-matrix and the cargo state sub-matrix and the feature fitting matrix is ​​calculated simultaneously using a dual-metric method.

[0157] Based on the fitness level, a corresponding candidate feature vector is generated, and the target steering parameters are generated based on the candidate feature vector.

[0158] Furthermore, the calculation module is specifically used for:

[0159] By combining the specific characteristics of the goods with the transportation constraints, corresponding evaluation factors are set, and the fit degree is dynamically weighted by the fuzzy hierarchical analysis algorithm. At the same time, the core feature dimension is extracted by the kernel principal component analysis algorithm to generate the corresponding candidate feature vector.

[0160] Based on vehicle dynamics rules, intermediate feature vectors that meet preset requirements are selected from the candidate feature vectors.

[0161] Based on a reinforcement learning framework, with the goals of minimizing cargo center of gravity shift and minimizing path tracking error, the optimal steering parameter mapping relationship is obtained by combining actual road condition feedback. Simultaneously, the intermediate feature vector is dynamically compensated and calibrated through the optimal steering parameter mapping relationship to generate the target steering parameters accordingly.

[0162] The fourth embodiment of the present invention provides a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the vehicle adaptive steering method as described above.

[0163] The fifth embodiment of the present invention provides a readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the vehicle adaptive steering method as described above.

[0164] In summary, the vehicle adaptive steering method and system provided by the above embodiments of the present invention can accurately control the vehicle's steering system based on driving data and actual transportation tasks, thereby improving the steering efficiency of the steering system.

[0165] It should be noted that the above modules can be functional modules or program modules, and can be implemented through software or hardware. For modules implemented through hardware, the above modules can reside in the same processor; or the above modules can be located in different processors in any combination.

[0166] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0167] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0168] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0169] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0170] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A vehicle adaptive steering method, characterized in that, The method includes: Collect historical driving big data from several vehicle models and several driving scenarios, extract common steering behavior features and personalized driving preference features under different driving scenarios through transfer learning networks, and simultaneously generate corresponding driving behavior feature sets; Collect the cargo characteristics, transportation constraints and safety level requirements of the current transportation task, quantify the task requirement vector using the entropy weight method, and simultaneously generate the corresponding feature adaptation matrix based on the driving behavior feature set and the task requirement vector. The vehicle's driving data and cargo status data are collected by the vehicle-mounted edge computing node, and the corresponding target steering parameters are generated simultaneously by combining the feature adaptation matrix. Based on the target steering parameters, a corresponding steering control command is generated, and the steering control command is simultaneously sent to the steering system to complete the corresponding steering control.

2. The vehicle adaptive steering method according to claim 1, characterized in that, The steps of collecting the cargo characteristics, transportation constraints, and safety level requirements of the current transportation task, and quantifying them using the entropy weight method to obtain the task requirement vector, include: The coupling relationship between the cargo characteristics, transportation constraints and safety level requirements is detected by the association rule mining algorithm, and composite constraint features are extracted based on the coupling relationship to create a corresponding task requirement index set. The initial information entropy weights corresponding to each indicator in the task requirement indicator set are calculated by the entropy weight method, and a dynamic adaptation factor for the transportation scenario is introduced to construct a mapping model between the adaptation factor and the indicator weights. The task requirement index set is dynamically modified using the mapping model to generate the corresponding task requirement vector.

3. The vehicle adaptive steering method according to claim 2, characterized in that, The step of dynamically correcting the task requirement index set through the mapping model to generate the corresponding task requirement vector includes: The correction parameters of each indicator output by the mapping model are used as evidence. A credibility allocation function is constructed by combining the fluctuation of cargo characteristics and the range of changes in transportation constraints. At the same time, each correction indicator in the task requirement indicator set is quantified through the credibility allocation function to generate the corresponding preliminary correction indicator set. The numerical and categorical indicators in the preliminary correction indicator set are transformed into multimodal feature vectors. Attention mechanism is used to calculate the correlation weights between the multimodal feature vectors to perform cross-dimensional optimization and generate a reinforcement correction indicator set. Using the enhanced correction index set as input and the historical best fit vector as a reference, the task requirement vector is trained accordingly through adversarial training of a generative adversarial network.

4. The vehicle adaptive steering method according to claim 1, characterized in that, The step of generating a corresponding feature adaptation matrix based on the driving behavior feature set and the task requirement vector includes: The driving behavior feature set and the indicators of each dimension of the task requirement vector are respectively used as graph nodes to construct a heterogeneous feature association graph. Simultaneously, the potential association strength between nodes is learned through a graph neural network to generate the corresponding graph feature matrix based on the output association representation. Using the graph feature matrix as input, a sparse Bayesian learning model is constructed, with the goal of minimizing the feature adaptation error, and the corresponding corrected feature matrix is ​​output. By introducing task requirement priority constraints, the modified feature matrix is ​​iteratively solved to generate the corresponding feature adaptation matrix.

5. The vehicle adaptive steering method according to claim 4, characterized in that, The step of introducing task requirement priority constraints and iteratively solving the modified feature matrix to generate the corresponding feature adaptation matrix includes: The Analytic Hierarchy Process (AHP) is introduced to construct a priority judgment matrix for task requirements. Simultaneously, priority quantification coefficients for each requirement dimension are calculated based on the priority judgment matrix, and the priority quantification coefficients are integrated into a corresponding priority constraint matrix. Using the product of the modified feature matrix and the priority constraint matrix as the initial input, an iterative solution framework based on the alternating direction multiplier method is constructed. In each iteration, Kalman filtering is introduced to dynamically estimate the solution bias in order to obtain the preliminary feature fitting matrix. Multiple sets of transportation scenario disturbance samples are generated using Monte Carlo simulation. The preliminary feature adaptation matrix is ​​then substituted into each sample for adaptation verification. Once the verification is successful, the corresponding feature adaptation matrix is ​​generated.

6. The vehicle adaptive steering method according to claim 1, characterized in that, The step of collecting vehicle driving data and cargo status data through on-board edge computing nodes and simultaneously generating corresponding target steering parameters by combining the feature adaptation matrix includes: The spatiotemporal alignment algorithm eliminates the data misalignment between the driving data and the cargo status data, and the isolated forest algorithm is used to filter abnormal data. The weight coefficients of the two types of data are dynamically allocated according to the characteristics of the cargo, and the corresponding standard data matrix is ​​generated synchronously. The standard data matrix is ​​decomposed into a driving state sub-matrix and a cargo state sub-matrix, and the degree of fit between the driving state sub-matrix and the cargo state sub-matrix and the feature fitting matrix is ​​calculated simultaneously using a dual-metric method. Based on the fitness level, a corresponding candidate feature vector is generated, and the target steering parameters are generated based on the candidate feature vector.

7. The vehicle adaptive steering method according to claim 6, characterized in that, The step of generating a corresponding candidate feature vector based on the fit, and generating the target steering parameters based on the candidate feature vector, includes: By combining the specific characteristics of the goods with the transportation constraints, corresponding evaluation factors are set, and the fit degree is dynamically weighted by the fuzzy hierarchical analysis algorithm. At the same time, the core feature dimension is extracted by the kernel principal component analysis algorithm to generate the corresponding candidate feature vector. Based on vehicle dynamics rules, intermediate feature vectors that meet preset requirements are selected from the candidate feature vectors. Based on a reinforcement learning framework, with the goals of minimizing cargo center of gravity shift and minimizing path tracking error, the optimal steering parameter mapping relationship is obtained by combining actual road condition feedback. Simultaneously, the intermediate feature vector is dynamically compensated and calibrated through the optimal steering parameter mapping relationship to generate the target steering parameters accordingly.

8. A vehicle adaptive steering system, characterized in that, The system includes: The data acquisition module is used to collect historical driving big data for several vehicle models and several driving scenarios. It extracts common steering behavior features and personalized driving preference features under different driving scenarios through a transfer learning network, and generates corresponding driving behavior feature sets in a synchronous manner. The generation module is used to collect the cargo characteristics, transportation constraints and safety level requirements of the current transportation task, quantify the task requirement vector using the entropy weight method, and simultaneously generate the corresponding feature adaptation matrix based on the driving behavior feature set and the task requirement vector. The computing module is used to collect vehicle driving data and cargo status data through the vehicle edge computing node, and simultaneously combine the feature adaptation matrix to generate the corresponding target steering parameters. The control module is used to generate corresponding steering control commands based on the target steering parameters, and synchronously send the steering control commands to the steering system to complete the corresponding steering control.

9. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the vehicle adaptive steering method as described in any one of claims 1 to 7.

10. A readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the vehicle adaptive steering method as described in any one of claims 1 to 7.

Images