Vehicle control methods and vehicles based on direction correction anomalies

Abstract

This application provides a vehicle control method and vehicle based on abnormal steering correction, relating to the field of autonomous driving technology. The method includes: first, determining the vehicle's temporal frequency characteristics, multi-dimensional driving behavior characteristics, and supplementary driving behavior characteristics based on multi-source vehicle operating data; second, fusing the temporal frequency characteristics, multi-dimensional driving behavior characteristics, and supplementary driving behavior characteristics to obtain fused decision features, which are used to characterize abnormal driver steering correction behavior; and finally, generating a vehicle control strategy based on the fused decision features, which is used to adjust the steering system's feedback rules and execution response methods for driver steering wheel operations in response to abnormal driver steering correction behavior. This solves the technical problem of low recognition accuracy of abnormal driver steering correction behavior in related technologies.

Description

Technical Field

[0001] This application relates to the field of autonomous driving technology, and in particular to a vehicle control method and vehicle based on direction correction anomalies. Background Technology

[0002] With the rapid iteration and mass production of vehicle intelligent technologies, steer-by-wire systems and advanced driver assistance systems have become mainstream configurations in passenger vehicles. The focus of active safety technology development has shifted from traditional passive accident protection to proactive risk prediction and refined intervention control during driving. Driver steering wheel corrections are the core component of vehicle lateral control. Frequent, unintentional, and sudden abnormal corrections are significant contributing factors to lane departure, vehicle yaw, and other driving safety risks. Therefore, identifying driver steering corrections and implementing corresponding adaptive steering system control have become core research and application directions in the field of active vehicle safety.

[0003] Most existing solutions for driver steering behavior recognition and vehicle control rely solely on single-dimensional operational data such as steering wheel angle and steering torque for behavior determination. These solutions lack sensitivity in capturing subtle, high-frequency directional corrections by the driver, resulting in a high rate of missed detections of abnormal behaviors. Some solutions that employ multi-source data fusion often use fixed-weight fusion strategies, which cannot adapt to the confidence fluctuations of data across different vehicle speeds and road conditions. They are also susceptible to environmental and sensor noise interference, leading to deviations in recognition results.

[0004] In summary, the relevant technologies suffer from the technical problem of low accuracy in recognizing abnormal driver directional correction behaviors. Summary of the Invention

[0005] In view of the above problems, this application provides a vehicle control method and vehicle based on direction correction anomalies that overcomes or at least partially solves the problem of low accuracy in recognizing abnormal driver direction correction behavior. The technical solution is as follows: A vehicle control method based on direction correction anomalies, the method comprising: Based on the vehicle's multi-source operation data, the vehicle's temporal frequency characteristics, multi-dimensional driving behavior characteristics, and supplementary driving behavior characteristics are determined. The temporal frequency characteristics are used to characterize the temporal characteristics of the driver's high-frequency and sudden directional correction behavior. The multi-dimensional driving behavior characteristics are used to characterize the directional correction behavior characteristics. The supplementary driving behavior characteristics are used to characterize the discrimination characteristics of abnormal directional correction behavior that is not predefined. The time-series frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features are fused to obtain fused decision features, which are used to characterize abnormal driver direction correction behavior. Based on the fused decision features, a vehicle control strategy is generated. The vehicle control strategy is used to correct abnormal behavior in response to driver direction and adjust the feedback rules and execution response mode of the steering system to the driver's steering wheel operation.

[0006] In this way, by simultaneously extracting three complementary features based on multi-source vehicle operation data—namely, the temporal characteristics of high-frequency sudden directional correction behavior, the overall characteristics of directional correction behavior, and the characteristics of undefined anomaly discrimination—full-dimensional coverage of typical and atypical abnormal directional correction behaviors of drivers is achieved, solving the problem of incomplete coverage of abnormal behavior identification by single-dimensional features. Furthermore, by fusing the three types of features, a fused decision feature capable of characterizing abnormal directional correction behavior of drivers is obtained. Compared with single-feature identification, this effectively reduces the probability of false positives and false negatives of abnormal behavior. Finally, based on the fused decision feature, a targeted vehicle control strategy is generated, directly adjusting the feedback rules and execution response methods of the steering system to the driver's steering wheel operation. This achieves precise adaptive intervention for abnormal directional correction behavior, effectively suppressing safety risks such as vehicle yaw instability and lane departure caused by abnormal correction behavior, while avoiding indiscriminate intervention that interferes with the driver's normal driving, thus balancing vehicle driving safety and driving smoothness.

[0007] Optionally, the temporal frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features are fused to obtain fused decision features, including: Determine the weights corresponding to the time-series frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features; The time-series frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features are fused based on the weights to obtain the fused decision features.

[0008] In this way, by first determining the weights corresponding to the three types of features, and then completing feature fusion based on the weights to obtain fused decision features, differentiated weighted fusion of different features is achieved. This can specifically strengthen the feature signals that effectively represent abnormal driver directional correction behavior, and weaken the interference of invalid or noisy features. Compared with unweighted feature splicing and equal-weighted fusion, this method effectively improves the accuracy of the fused decision features in representing abnormal directional correction behavior, reduces the recognition bias caused by feature redundancy, and provides a more reliable and focused decision basis for the subsequent generation of vehicle control strategies.

[0009] Optionally, determining the weights corresponding to the time-series frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features includes: After aligning the temporal frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features to the same time reference and feature dimension, the temporal change trend matching degree of each pair of features within the same temporal window is determined to obtain the temporal response consistency index. Extract the direction correction behavior representation semantic vectors corresponding to the temporal frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features respectively, determine the semantic vector space matching degree of each pair of features, and obtain the behavior semantic consistency index. Based on the temporal response consistency index and the behavioral semantic consistency index, the weights corresponding to the temporal frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features are determined.

[0010] This approach first aligns the three types of features to the same time reference and feature dimension, eliminating computational biases caused by temporal misalignment and dimensional mismatch of different features, thus ensuring the accuracy of subsequent consistency index calculations. Then, by calculating the temporal trend matching degree and semantic vector space matching degree of each pair of features, temporal response consistency index and behavioral semantic consistency index are obtained respectively. This allows for cross-validation of the consistency of the three types of features in representing the driver's directional correction actions from both temporal and semantic dimensions. Finally, feature weights are determined based on the dual-dimensional consistency index, allowing features with more stable representations and higher matching degrees of directional correction behavior to receive higher weights. This significantly reduces noise interference from low-consistency features, making the weight allocation more closely aligned with the actual representational capabilities of the features. This further enhances the reliability of the fused decision features in identifying abnormal driver directional correction behavior, reducing misjudgments and missed judgments of abnormal behaviors.

[0011] Optionally, determining the weights corresponding to the temporal frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features based on the temporal response consistency index and the behavioral semantic consistency index includes: Based on the temporal response consistency index and the behavioral semantic consistency index, the confidence scores corresponding to the temporal frequency features, the multi-dimensional driving behavior features and the supplementary driving behavior features are determined respectively. The confidence score is normalized to determine the contribution of the time-series frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features to the driver's direction correction behavior in the process of identifying abnormal driver direction correction behavior. Based on the contribution, the weights corresponding to the time-series frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features are determined.

[0012] In this way, by converting the temporal response consistency index and the behavioral semantic consistency index into a comprehensive confidence score of the corresponding features, a quantitative assessment of the confidence of a single feature in representing the driver's directional correction behavior is achieved. Then, by normalizing the comprehensive confidence score, the contribution of each feature in the process of identifying directional correction behavior anomalies is obtained. Finally, feature weights are determined based on the contribution, realizing a direct binding between weights and the actual contribution of features to anomaly identification. This completes the quantifiable allocation of weights, avoids the bias caused by subjective weight settings, and ensures that the fused decision features can focus on the most effective feature signals for identifying directional correction anomalies, further improving the accuracy of identifying abnormal driver directional correction behavior and the reliability of control decisions.

[0013] Optionally, generating a vehicle control strategy based on the fused decision features includes: The fused decision features are mapped to the vehicle lateral control parameter space to obtain adjustment parameters for adjusting the response characteristics of the vehicle steering system; The vehicle control strategy is generated based on the adjusted parameters.

[0014] In this way, by mapping the fusion decision features representing abnormal driver directional correction behavior to the vehicle's lateral control parameter space, the abstract abnormal behavior representation is transformed into adjustment parameters that can be directly applied to adjust the steering system's response characteristics. Then, a vehicle control strategy is generated based on the adjustment parameters. This achieves a direct transformation from anomaly identification results to control execution actions, allowing the generated control strategy to match the currently detected abnormal directional correction behavior. This ensures the targeted nature of control intervention and avoids indiscriminate intervention that interferes with the driver's normal driving. At the same time, by adjusting the steering system's response characteristics, abnormal driver directional correction behavior can be suppressed, effectively reducing the safety risks of lateral vehicle movement.

[0015] Optionally, generating the vehicle control strategy based on the adjustment parameters includes: Obtain vehicle operating condition data; The adjustment parameters are corrected based on the driving condition data to obtain the target control parameters; The vehicle control strategy is generated based on the target control parameters.

[0016] In this way, by introducing real-time vehicle driving condition data to correct the steering system adjustment parameters during the generation of vehicle control strategy, and then generating the control strategy after obtaining the target control parameters adapted to the current driving conditions, the final control strategy can be adapted to the current driving scenario of the vehicle. This solves the problem that fixed adjustment parameters are prone to over-intervention or under-intervention under different driving conditions. It realizes appropriate intervention in abnormal driver steering correction behavior under different driving scenarios such as vehicle speed and road environment. This can effectively avoid driving safety risks caused by abnormal correction behavior, and ensure the driver's driving experience and the smoothness of vehicle handling.

[0017] Optionally, the method further includes: After the vehicle control strategy is executed, feedback data on vehicle status and driver operation are acquired. Based on the feedback data, the extraction rules and fusion weights corresponding to the time-series frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features are optimized.

[0018] In this way, after the vehicle control strategy is executed, by collecting feedback data on vehicle status and driver operation, the identification and control effect of abnormal driver directional correction behavior can be verified. Then, based on the feedback data, the extraction rules and fusion weights of the three types of features are optimized, and a complete closed loop of "abnormal identification - control execution - effect feedback - iterative optimization" is constructed. This allows the feature extraction and fusion logic to continuously adapt to changes in driver driving habits and different driving scenarios, continuously improving the identification accuracy and control intervention effectiveness of abnormal driver directional correction behavior, and solving the problem of insufficient long-term adaptability caused by fixed identification and fusion logic.

[0019] Optionally, optimizing the extraction rules and fusion weights corresponding to the time-series frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features based on the feedback data includes: The feedback data is integrated with the multi-source operational data to obtain an optimized driving behavior sample; Based on the optimized driving behavior sample, the feature values ​​corresponding to the time-series frequency feature, the multi-dimensional driving behavior feature, and the supplementary driving behavior feature are determined using the extraction rules. Determine the deviation between the feature value and the sample value; The extraction rules are optimized with the goal of minimizing the aforementioned deviation. With the improvement of the confidence level of the fused decision features as the optimization objective, the fusion weights are updated based on the optimized driving behavior samples.

[0020] In this way, by integrating feedback data with original multi-source operational data, optimized driving behavior samples with real control effect verification are obtained, providing a reliable supervisory basis for optimizing feature extraction rules and fusion weights. By calculating the deviation between feature output values ​​and sample values, and optimizing the extraction rules with the goal of minimizing the deviation, the representation accuracy of the three types of features for driver directional correction behavior can be directly improved, reducing the bias of anomaly identification. At the same time, updating the fusion weights with the goal of improving the confidence of fusion decision features can continuously optimize the effect of feature fusion, enabling the entire anomaly identification and fusion system to achieve continuous iteration and continuously improve the accuracy of identifying abnormal driver directional correction behavior and the reliability of control decisions.

[0021] Optionally, determining the vehicle's temporal frequency characteristics, multi-dimensional driving behavior characteristics, and supplementary driving behavior characteristics based on the vehicle's multi-source operational data includes: After preprocessing the multi-source operation data, the temporal variation pattern of the driver's direction correction actions, direction correction behavior, and the discrimination characteristics of predefined abnormal direction correction behavior are extracted from the multi-source operation data through a preset feature extraction strategy. Based on the temporal variation pattern of the driver's directional correction actions, the directional correction behavior, and the discrimination characteristics of predefined abnormal directional correction behaviors, the temporal frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features are determined respectively.

[0022] In this way, by preprocessing multi-source operational data, problems such as noise and temporal misalignment in the original data are effectively eliminated, providing a high-quality and standardized input foundation for feature extraction. Then, through a preset feature extraction strategy, the temporal variation patterns, behavioral characteristics, and predefined anomaly discrimination characteristics directly related to the driver's directional correction behavior are extracted. Three types of features are then determined respectively, so that each type of feature corresponds to a core dimension of directional correction behavior, ensuring the targeting and effectiveness of feature extraction, avoiding interference from irrelevant data on feature representation, and providing a reliable feature foundation for subsequent directional correction anomaly identification and vehicle control strategy generation.

[0023] A vehicle control device based on direction correction anomalies, the device comprising: The determination module is used to determine the vehicle's temporal frequency characteristics, multi-dimensional driving behavior characteristics, and supplementary driving behavior characteristics based on the vehicle's multi-source operating data. The temporal frequency characteristics are used to characterize the temporal characteristics of the driver's high-frequency and sudden directional correction behaviors. The multi-dimensional driving behavior characteristics are used to characterize the directional correction behavior characteristics. The supplementary driving behavior characteristics are used to characterize the discrimination characteristics of abnormal directional correction behaviors that are not predefined. The fusion module is used to fuse the time-series frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features to obtain fused decision features, which are used to characterize abnormal driver direction correction behavior. The generation module is used to generate a vehicle control strategy based on the fused decision features. The vehicle control strategy is used to correct abnormal behavior in response to driver direction and adjust the feedback rules and execution response mode of the steering system to the driver's steering wheel operation.

[0024] Optionally, the fusion module is also used for: Determine the weights corresponding to the time-series frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features; The time-series frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features are fused based on the weights to obtain the fused decision features.

[0025] Optionally, the fusion module is also used for: After aligning the temporal frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features to the same time reference and feature dimension, the temporal change trend matching degree of each pair of features within the same temporal window is determined to obtain the temporal response consistency index. Extract the direction correction behavior representation semantic vectors corresponding to the time-series frequency features, the time-multidimensional driving behavior features, and the supplementary driving behavior features respectively, determine the semantic vector space matching degree of each pair of features, and obtain the behavior semantic consistency index. Based on the temporal response consistency index and the behavioral semantic consistency index, the weights corresponding to the temporal frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features are determined.

[0026] Optionally, the fusion module is also used for: Based on the temporal response consistency index and the behavioral semantic consistency index, the confidence scores corresponding to the temporal frequency features, the multi-dimensional driving behavior features and the supplementary driving behavior features are determined respectively. The confidence score is normalized to determine the contribution of the time-series frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features to the driver's direction correction behavior in the process of identifying abnormal driver direction correction behavior. Based on the contribution, the weights corresponding to the time-series frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features are determined.

[0027] Optionally, the generation module is also used for: The fused decision features are mapped to the vehicle lateral control parameter space to obtain adjustment parameters for adjusting the response characteristics of the vehicle steering system; The vehicle control strategy is generated based on the adjusted parameters.

[0028] Optionally, the generation module is also used for: Obtain vehicle operating condition data; The adjustment parameters are corrected based on the driving condition data to obtain the target control parameters; The vehicle control strategy is generated based on the target control parameters.

[0029] Optionally, the vehicle control device based on the direction correction anomaly also includes an acquisition module: The acquisition module is used to acquire feedback data on vehicle status and driver operation after the vehicle control strategy is executed. Based on the feedback data, the extraction rules and fusion weights corresponding to the time-series frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features are optimized.

[0030] Optionally, the acquisition module is also used for: The feedback data is integrated with the multi-source operational data to obtain an optimized driving behavior sample; Based on the optimized driving behavior sample, the feature values ​​corresponding to the time-series frequency feature, the multi-dimensional driving behavior feature, and the supplementary driving behavior feature are determined using the extraction rules. Determine the deviation between the feature value and the sample value; The extraction rules are optimized with the goal of minimizing the aforementioned deviation. With the improvement of the confidence level of the fused decision features as the optimization objective, the fusion weights are updated based on the optimized driving behavior samples.

[0031] Optionally, the determining module is also used for: After preprocessing the multi-source operation data, the temporal variation pattern of the driver's direction correction actions, direction correction behavior, and the discrimination characteristics of predefined abnormal direction correction behavior are extracted from the multi-source operation data through a preset feature extraction strategy. Based on the temporal variation pattern of the driver's directional correction actions, the directional correction behavior, and the discrimination characteristics of predefined abnormal directional correction behaviors, the temporal frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features are determined respectively.

[0032] A vehicle comprising: the vehicle performing any of the optional vehicle control methods based on direction correction anomalies described above.

[0033] By employing the aforementioned technical solution, this application provides a vehicle control method based on anomaly-based steering correction. Using multi-source vehicle operation data as the basis for identification, compared to single-dimensional steering data, it provides more comprehensive and multi-dimensional data source support for identifying driver steering correction behavior. Based on this, three types of complementary and clearly defined features are specifically designed and extracted. The combined design of these three types of features achieves full-scene and full-dimensional coverage of typical and atypical, transient and steady-state steering correction anomalies, improving the completeness and accuracy of anomaly behavior capture from the source. Furthermore, by fusing the three complementary features, a fused decision feature is obtained. Compared to identification results relying on a single feature, it can effectively avoid single-feature errors through cross-validation of multi-dimensional features. The feature is susceptible to data noise and has a large identification bias. This feature significantly reduces the probability of false positives and false negatives of abnormal behavior, providing a high-confidence and high-focus decision basis for the generation of subsequent control strategies. Finally, based on the fused decision features, a targeted vehicle control strategy is generated. For the detected abnormal driver directional correction behavior, the feedback rules and execution response mode of the steering system to the driver's steering wheel operation are directly adjusted. This achieves a direct closed loop from the abnormal behavior identification result to the steering control execution action, so that the intervention action of the steering system matches the attributes of the currently detected abnormal behavior. Through the targeted adjustment of steering feedback rules and execution response mode, the driving safety risks such as vehicle yaw instability and lane departure caused by the driver's unintentional abnormal directional correction behavior can be effectively suppressed.

[0034] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0035] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 One of the flowcharts of the vehicle control method based on direction correction anomaly provided in this application is shown; Figure 2 The second schematic flowchart of the vehicle control method based on direction correction anomaly provided in this application embodiment is shown; Figure 3 The third schematic flowchart of the vehicle control method based on direction correction anomaly provided in this application is shown; Figure 4The fourth schematic flowchart of the vehicle control method based on direction correction anomaly provided in this application is shown. Figure 5 A schematic diagram of a vehicle control device based on a direction correction anomaly provided in an embodiment of this application is shown. Detailed Implementation

[0036] Exemplary embodiments of the present application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thorough and complete, and will fully convey the scope of the present application to those skilled in the art.

[0037] To address the technical problem of low accuracy in recognizing abnormal driver direction correction behavior in related technologies, this application provides a vehicle control method based on abnormal direction correction, such as... Figure 1 As shown, Figure 1 This is a schematic flowchart illustrating a vehicle control method based on direction correction anomalies provided in an embodiment of this application. The method includes: S11. Based on the multi-source operation data of the vehicle, determine the vehicle's temporal frequency characteristics, multi-dimensional driving behavior characteristics, and supplementary driving behavior characteristics.

[0038] Among them, the temporal frequency feature is used to characterize the temporal characteristics of the driver's high-frequency and sudden directional correction behavior, the multi-dimensional driving behavior feature is used to characterize the characteristics of directional correction behavior, and the supplementary driving behavior feature is used to characterize the discriminative characteristics of abnormal directional correction behavior that is not predefined.

[0039] In one specific embodiment, after preprocessing the multi-source operation data, the temporal variation pattern of the driver's directional correction actions, directional correction behavior, and the discrimination characteristics of predefined abnormal directional correction behaviors are extracted from the multi-source operation data through a preset feature extraction strategy. Based on the temporal variation pattern of the driver's directional correction actions, directional correction behavior, and the discrimination characteristics of predefined abnormal directional correction behaviors, temporal frequency features, multi-dimensional driving behavior features, and supplementary driving behavior features are determined respectively.

[0040] Specifically, multi-source vehicle operational data is acquired. This multi-source operational data includes at least one or more of the following: driver operation data, vehicle motion state data, surrounding environment data, and driver behavior state data. Examples include steering wheel angle / speed signals, yaw rate, vehicle longitudinal speed, lane departure, driver hand position data, in-vehicle camera behavioral images, and meteorological environmental level parameters. The acquired multi-source operational data is preprocessed. Preprocessing steps include at least timestamp alignment, filtering and noise reduction, and dimensionality normalization. The aim is to eliminate noise interference, temporal misalignment, and dimensional mismatch in the original data, resulting in standardized, normalized data. High-quality feature extraction input; based on preprocessed multi-source operational data, and through a preset feature extraction strategy, three core features directly related to the driver's directional correction behavior are extracted: the temporal variation pattern of the driver's directional correction actions, the overall characteristics of directional correction behavior, and the discriminative characteristics of abnormal directional correction behavior not predefined. The core logic of this extraction strategy is "one-to-one correspondence between features and characteristics," ensuring that the extraction of each type of feature has a clear target and basis. Finally, the three types of extracted features are respectively transformed into temporal frequency features, multi-dimensional driving behavior features, and supplementary driving behavior features, thus completing the determination of the three types of features.

[0041] For example, multi-source operational data during vehicle operation is collected through sensors, specifically including: driver steering wheel angle data (sampling frequency 100Hz), steering torque data (sampling frequency 100Hz), real-time vehicle speed data (sampling frequency 50Hz), lane departure data (sampling frequency 20Hz), and driver blink frequency data (sampling frequency 10Hz). All the above data are timestamped. A Gaussian filtering algorithm is used to filter and denoise the steering wheel angle and steering torque data, eliminating noise interference caused by road bumps and sensor errors. Finally, all data undergoes dimensionality normalization to eliminate extraction bias caused by differences in data magnitude. Through a preset feature extraction strategy, three types of characteristics are extracted: ① A "sparse gating temporal modeling strategy" is adopted, introducing a sparse attention weight mechanism. A 1-second sliding window (moving step size 200ms) captures high-frequency fluctuation signals of 1-5Hz, focusing on extracting parameters such as the frequency of steering wheel reversal actions, fluctuation amplitude, and fluctuation duration, forming a feature vector characterizing the temporal characteristics of the driver's high-frequency, sudden direction correction behavior (as defined in claim 1); within the sliding window, the following conditions are met: "more than three reversal actions +..." Sequences with a single correction amplitude exceeding 2 degrees are marked as high-frequency abnormal correlation feature segments; ② A "dual-domain perception fusion strategy" is adopted to simultaneously extract time-domain statistical features (turning angle amplitude, peak torque, steering duration, steering rate) and frequency-domain energy distribution features (obtained through STFT (Short-Time Fourier Transform) short-time Fourier analysis) to construct a feature vector aligned with both angular velocity and frequency domains, comprehensively characterizing the characteristics of directional correction behavior, while integrating contextual parameters such as lane offset and vehicle speed to strengthen the correlation between features and driving scenarios; ③ A "zero-sample semantic discrimination strategy" is adopted to extract behavioral attribute vectors (correction rhythm density, left-right switching asymmetry, peak angular velocity) based on predefined abnormal behavior semantic boundaries, and capture discrimination characteristics that differ from known abnormal semantic vectors through comparative memory metric learning to form feature vectors representing undefined abnormal behaviors; for extreme scenarios where data cannot be collected, abnormal samples are supplemented and synthesized through a perturbation data generation mechanism to improve feature generalization discrimination ability.

[0042] In this embodiment, based on multi-source vehicle operation data, three types of complementary and clearly positioned features are extracted in a targeted manner. This solves the core pain points of related technologies, such as incomplete coverage, insufficient sensitivity, and weak generalization ability of single features in recognizing driver directional correction behavior. It achieves full-scene and full-dimensional capture of typical and atypical, high-frequency sudden and regular steady-state directional correction behaviors of drivers.

[0043] S12. The temporal frequency features, multi-dimensional driving behavior features, and supplementary driving behavior features are fused to obtain the fused decision features.

[0044] Among them, the fused decision features are used to characterize abnormal driver directional correction behavior.

[0045] Specifically, the temporal frequency features, multi-dimensional driving behavior features, and supplementary driving behavior features determined in S11 are adapted to ensure that the three types of features have a unified temporal benchmark and feature dimension specifications, avoiding fusion deviations caused by temporal misalignment and dimension mismatch. The core of the adaptation process is to unify the vector specifications of the three types of features based on the unified temporal benchmark preprocessed in S11. A preset feature fusion strategy is then used to fuse the adapted three types of features. The core logic of the fusion strategy is "differentiated fusion based on feature representation capabilities," ensuring that features that contribute highly to anomaly identification receive higher weights, thereby improving the accuracy of the fusion results. Finally, through fusion processing, the representational information of the three types of features is integrated into a single fusion decision feature. The core function of this fusion decision feature is to characterize the abnormal attributes of the driver's directional correction behavior.

[0046] For example, based on the unified millisecond-level time series benchmark in S11, the vector specifications of the three types of features are unified into a 5-dimensional feature vector (adapted to the core dimensions of time-series frequency features, multi-dimensional driving behavior features, and supplementary driving behavior features). The time-series frequency feature vector consists of [high-frequency fluctuation amplitude, fluctuation frequency, fluctuation duration, fluctuation interval, and fluctuation peak value], the multi-dimensional driving behavior feature vector consists of [turning angle amplitude, torque peak value, steering duration, steering rate, and steering smoothness], and the supplementary driving behavior feature vector consists of [semantic difference value, anomaly confidence, anomaly duration, anomaly frequency, and anomaly deviation], ensuring the three types of features are fusionable. A weighted fusion strategy is adopted, combining the representational capabilities of the three types of features, to initially determine the fusion weight (time-series frequency feature weight 0.35). The core logic of the weight allocation is as follows: multi-dimensional driving behavior features have the highest weight because they are the most accurate in representing conventional anomalies; time-series frequency features are the most sensitive in capturing high-frequency sudden anomalies, so their weight is second; and supplementary driving behavior features are used for generalization and have a relatively low weight. Based on this weight, the three types of feature vectors are weighted and summed to obtain a fusion decision feature vector. The fusion decision feature vector is [0.82, 0.75, 0.33, 0.68, 0.21]. By judging through a preset threshold (anomaly threshold of 0.5), the anomaly confidence level corresponding to this fusion decision feature vector is 0.75, which is greater than the anomaly threshold. This clearly indicates that the driver has abnormal directional correction behavior, and the anomaly level is moderate.

[0047] In this embodiment, the temporal frequency features, multi-dimensional driving behavior features, and supplementary driving behavior features extracted in S11 are fused to solve the defects of single feature representation being incomplete, susceptible to interference, and having low recognition accuracy, and to achieve synergistic effect of multi-dimensional features. The driver's directional correction behavior is characterized from three dimensions: high-frequency sudden anomaly, regular anomaly, and atypical anomaly. By fusing, the advantages of various features can be integrated, and the noise interference and recognition bias of single features can be offset by cross-validation, finally obtaining a fused decision feature that can comprehensively represent the abnormality of the driver's directional correction behavior.

[0048] S13. Generate vehicle control strategies based on fused decision features.

[0049] Among them, the vehicle control strategy is used to correct abnormal behavior in response to driver direction, and to adjust the feedback rules and execution response of the steering system to the driver's steering wheel operation.

[0050] Specifically, the fused decision features obtained in S12 are analyzed to clarify the abnormal attributes of the driver's directional correction behavior, including at least the presence, type, and level of the abnormality. Based on the analyzed abnormal attributes, the fused decision features are mapped to the vehicle steering control parameter space to obtain control parameters for adjusting the steering system's response characteristics. The core function of these control parameters is to clarify "what to adjust and how much to adjust" for the steering system, providing a quantitative basis for the generation of the control strategy. Finally, based on the mapped control parameters and combined with the control logic of the vehicle steering system, a complete vehicle control strategy is generated. The core of this control strategy is to adjust the steering system's feedback rules and execution response methods to the driver's steering wheel operations, ensuring that the control strategy can adapt to the current abnormal behavior and achieve effective intervention control.

[0051] For example, the fused decision feature vector [0.82, 0.75, 0.33, 0.68, 0.21] obtained in S12 is analyzed to determine the abnormal attributes of the driver's directional correction behavior: presence or absence of abnormality (abnormality exists), abnormality category (high-frequency sudden abnormality, due to the fluctuation amplitude and frequency of the time-series frequency feature exceeding the standard), and abnormality level (moderate, abnormality confidence of 0.75 in the range of 0.6-0.8). Based on the analyzed moderate high-frequency sudden abnormality, the fused decision features are mapped to the vehicle steering control parameter space to obtain the corresponding adjustment parameters: the steering feedback damping coefficient is adjusted to 1.2 (default value 1.0, increasing damping to suppress high-frequency operation), and the steering assist gain coefficient is adjusted to 0.8 (default value). Version 1.0 reduces power assist to mitigate steering response, and adjusts steering response lag time to 50ms (default 30ms, extended lag to suppress sudden maneuvers). Based on these adjustments, and combined with the vehicle steering system's control logic, a complete vehicle control strategy is generated, specifically: ① Feedback rule adjustment: When the driver performs high-frequency, sudden steering wheel corrections, the steering system automatically increases steering wheel torque feedback damping, allowing the driver to feel significant operational resistance and guiding the driver to reduce the frequency of operations; ② Execution response mode adjustment: Reduces steering assist gain and extends steering response lag time to prevent oversteering caused by sudden driver corrections, while ensuring the driver can still control the steering wheel and not deprive them of driving control.

[0052] In this embodiment, based on the fused decision features obtained in S12, a targeted vehicle control strategy is generated, realizing a direct closed loop of "anomaly identification → control intervention". This solves the defects of anomaly identification and control intervention being disconnected, intervention being untargeted, and poor adaptability in related technologies. This vehicle control strategy is specifically designed for abnormal driver directional correction behavior. By adjusting the feedback rules and execution response mode of the steering system to the driver's steering wheel operation, it can suppress the driver's unintentional abnormal directional correction behavior and effectively avoid driving safety risks such as vehicle yaw instability and lane departure.

[0053] The above scheme uses multi-source vehicle operation data as the basis for identification. Compared with single-dimensional steering data, it provides more comprehensive and multi-dimensional data source support for identifying driver directional correction behavior. Based on this, three types of complementary and clearly defined features are designed and extracted. The combination of these three types of features achieves full-scene and full-dimensional coverage of typical and atypical, transient and steady-state abnormal directional correction behaviors of drivers, improving the completeness and accuracy of abnormal behavior capture from the source. Furthermore, by fusing the three complementary features, a fused decision feature is obtained. Compared with identification results relying on a single feature, the cross-validation of multi-dimensional features effectively avoids the susceptibility of single features to data noise interference. By eliminating the large deviation, the probability of false positives and false negatives in abnormal behavior is significantly reduced, providing a high-confidence and highly focused decision-making basis for the generation of subsequent control strategies. Finally, based on the fused decision features, a targeted vehicle control strategy is generated. For the detected abnormal behavior of driver direction correction, the feedback rules and execution response mode of the steering system to the driver's steering wheel operation are directly adjusted, realizing a direct closed loop from abnormal behavior identification results to steering control execution actions. This allows the intervention action of the steering system to match the attributes of the currently detected abnormal behavior. Through the targeted adjustment of steering feedback rules and execution response mode, the driving safety risks such as vehicle yaw instability and lane departure caused by the driver's unintentional abnormal direction correction behavior can be effectively suppressed.

[0054] In some embodiments, such as Figure 2 As shown, time-series frequency features, multi-dimensional driving behavior features, and supplementary driving behavior features are fused to obtain fused decision features, including: S121. Determine the weights corresponding to the time-series frequency features, multi-dimensional driving behavior features, and supplementary driving behavior features.

[0055] Specifically, after aligning the temporal frequency features, multi-dimensional driving behavior features, and supplementary driving behavior features to the same time reference and feature dimension, the matching degree of the temporal change trend of each pair of features within the same time window is determined, resulting in a temporal response consistency index. This involves adapting the temporal frequency features, multi-dimensional driving behavior features, and supplementary driving behavior features determined in S11, with the core being to eliminate differences in the temporal reference and feature dimension among the three types of features, ensuring that the three types of features are fusionable and quantifiable. Secondly, within the same time window, each pair of features is calculated separately. The temporal trend matching degree of driver's directional correction actions reflects the consistency of the temporal representation of sudden / routine correction actions by the three types of features from the time dimension, avoiding weight bias caused by temporal misalignment. Semantic vectors representing directional correction actions corresponding to temporal frequency features, multi-dimensional driving behavior features, and supplementary driving behavior features are extracted respectively. The semantic vector space matching degree of each pair of features is determined to obtain the behavioral semantic consistency index. The behavioral semantic consistency index reflects the unity of the representation of directional correction behavior attributes by the three types of features from the semantic dimension, avoiding weight distortion caused by semantic representation conflict.

[0056] Specifically, based on the temporal response consistency index and the behavioral semantic consistency index, the comprehensive confidence scores corresponding to temporal frequency features, multi-dimensional driving behavior features, and supplementary driving behavior features are determined respectively. This comprehensive confidence score quantifies the comprehensive representation ability of a single type of feature. The comprehensive confidence scores are normalized to eliminate the differences in the dimensions of different indicators and are transformed into the contribution of the three types of features in anomaly identification, realizing the direct binding between weights and actual contributions. The contributions of temporal frequency features, multi-dimensional driving behavior features, and supplementary driving behavior features to the driver's directional correction behavior in the process of anomaly identification are determined respectively. Finally, based on the contribution, the weights corresponding to temporal frequency features, multi-dimensional driving behavior features, and supplementary driving behavior features are determined.

[0057] For example, based on the millisecond-level time series benchmark determined in the S11 preprocessing stage, the three types of feature vectors are unified to a preset 5-dimensional specification through a feature alignment model. A time series feature matching model is used, selecting a 200ms identical time series window centered on the trigger time of the driver's directional correction action, and inputting the aligned three types of feature vectors. The model learns the temporal variation pattern of the driver's directional correction action, calculates the temporal variation trend matching degree between pairs of features, and outputs the matching degree results of each pair of features. Finally, the model's built-in mean calculation module averages all pairwise matching degree results to obtain a temporal response consistency index. A pre-trained behavioral semantic encoding model is used to semantically encode the aligned three types of feature vectors, extracting the directional correction behavior representation semantic vector (a low-dimensional dense vector used to represent the behavioral attributes corresponding to the feature) for each type of feature. Subsequently, a semantic vector matching model is used to calculate the spatial matching degree of each pair of feature semantic vectors, quantifying the consistency of the three types of features in behavioral semantic representation. Finally, the model's mean calculation module is used to perform pairwise semantic matching. The matching results are averaged to obtain the behavioral semantic consistency index. A feature confidence assessment model is used, inputting the temporal response consistency index and the behavioral semantic consistency index. The model weights and fuses the two indices according to a preset weight allocation rule (60% for temporal response consistency index and 40% for behavioral semantic consistency index), outputting the comprehensive confidence scores for the three types of features, thus achieving model-based quantification of the comprehensive representation capability of a single type of feature. A normalization processing model is used to perform min-max normalization on the comprehensive confidence scores of the three types of features, eliminating the dimensional differences between different indices and mapping the scores to the [0,1] interval. This outputs the contribution of each type of feature in the driver's directional correction anomaly identification process, thus binding the contribution to the feature representation capability. Based on the weight allocation model, the contribution of the three types of features is input. Following the core logic of "positive correlation between contribution and weight," the model automatically allocates the weights corresponding to each type of feature, outputting the final feature weights (weights for temporal frequency features, multi-dimensional driving behavior features, and supplementary driving behavior features), completing the weight determination.

[0058] In this embodiment, a full-process weight determination method consisting of temporal benchmark alignment, dual consistency index (temporal response + behavioral semantics) quantification, confidence normalization, and contribution binding solves the defects of subjective and one-sided feature weight allocation in related technologies. It quantifies the true representation ability of three types of features for abnormal driver directional correction behavior from two dimensions: temporal matching (time dimension) and semantic representation consistency (semantic dimension), and finally strongly binds the weights to the actual contribution of the features. This ensures the objectivity and scientific nature of the weight allocation, highlights the dominant role of features with high contribution to anomaly identification (such as multi-dimensional features that comprehensively represent normal behavior and temporal features that capture sudden anomalies), and weakens the interference of feature noise with low consistency and low representation ability.

[0059] S122. Based on weights, the time-series frequency features, multi-dimensional driving behavior features, and supplementary driving behavior features are fused to obtain fused decision features.

[0060] Specifically, the time-series frequency features, multi-dimensional driving behavior features, and supplementary driving behavior features that have achieved "alignment with the same time reference and feature dimension" in S121 are used as fusion inputs to ensure the consistency of the input feature specifications; based on the three types of feature weights determined in S121, a preset weighted fusion strategy is adopted to perform weighted operations on the three types of feature vectors, and integrate the representation information of the three types of features into a single feature vector; the weighted fused feature vector is used as the final fusion decision feature.

[0061] For example, the aligned three types of feature vectors and the dynamic weights determined by S121 are simultaneously input into the feature weighted fusion model. The feature weighted fusion model first verifies the validity of the input feature vectors and weights to ensure that the feature specifications and weight allocation match (avoiding fusion deviation due to input anomalies). After the verification is passed, it enters the fusion operation stage. Through a three-way dynamic weighted integrator, the sparse activation sequence of time-series frequency features, the dual-domain tensor of multi-dimensional driving behavior features, and the semantic vector of supplementary driving behavior features are weighted and summed. During the fusion process, the representation role of high-weight features (features that contribute highly to anomaly identification) is strengthened, while the noise interference of low-weight features is weakened, achieving synergistic effect of the three types of features. Finally, the fusion decision feature vector is output. This vector has a preset 5-dimensional specification, including anomaly confidence, anomaly type (distinguishing between high-frequency sudden anomalies / regular anomalies / atypical anomalies), and Zero-Shot. The fuzzy classification identifier is used, where the anomaly confidence is calculated by the cumulative probability of anomalies within a sliding window. The fused decision feature vector is input into the anomaly discrimination model. The anomaly discrimination model analyzes the anomaly confidence, anomaly type and other information in the fused decision features through preset anomaly thresholds and discrimination rules, and outputs the anomaly judgment result of the driver's directional correction behavior (such as "there is a moderate high-frequency sudden directional correction anomaly"). This result can be directly used to generate the S13 vehicle control strategy.

[0062] In this embodiment, the weights determined in S121 are used as the core to fuse the three types of features, achieving the goal of "differentiated feature integration + anomaly representation". Unlike traditional methods such as unweighted splicing and fixed-weight fusion, the three types of features are integrated by weighting them with dynamic weights. This amplifies the representational role of features that contribute highly to the identification of direction correction anomalies (time-series frequency features and multi-dimensional driving behavior features), while retaining the value of supplementary driving behavior features in distinguishing atypical anomalies. The resulting fused decision features can comprehensively represent the abnormal attributes of the driver's direction correction behavior from three dimensions: high-frequency bursts, regular attributes, and atypical discrimination. This effectively offsets the noise interference and representation bias of single features, and significantly reduces the false detection rate and false negative rate of abnormal behavior.

[0063] In the above scheme, the comprehensive score and contribution of feature confidence are obtained based on the temporal response consistency index and the behavioral semantic consistency index. This objectively determines the fusion weight of the three types of features. Based on this weight, the temporal frequency features, multi-dimensional driving behavior features and supplementary driving behavior features are weighted and fused. This eliminates the subjectivity and one-sidedness of fixed weight allocation, fully highlights the actual contribution of various features to the identification of direction correction anomalies, strengthens the role of features with high representational ability and weakens noise interference during the fusion process, so that the final fused decision features can comprehensively reflect the abnormal attributes of the driver's direction correction behavior, significantly reduce the probability of false detection and false negative detection of anomalies, and provide a stable and reliable decision basis for the subsequent generation of vehicle control strategies.

[0064] In some embodiments, such as Figure 3 As shown, a vehicle control strategy is generated based on fused decision features, including: S131. Map the fused decision features to the vehicle lateral control parameter space to obtain adjustment parameters for adjusting the response characteristics of the vehicle steering system.

[0065] Specifically, the validity of the fused decision features output by S122 is verified to confirm the completeness and validity of the dimensions and anomaly representation information of the fused decision features, eliminate mapping deviations caused by invalid features, and ensure that the input meets the mapping requirements. Based on the preset feature-parameter mapping model, the valid fused decision features are mapped to the vehicle lateral control parameter space (this space contains all adjustable parameters related to the steering system response characteristics and is the core parameter set of steering system control, such as steering response bandwidth, EPS gain coefficient (Electric Power Steering Gain Coefficient), steering feedback damping coefficient, steering response hysteresis time, and other control parameters). The core logic of the mapping is "one-to-one correspondence between anomaly attributes and control parameters", that is, the anomaly type and anomaly level represented by the fused decision features are mapped to the specific adjustment direction and adjustment magnitude of the steering system response characteristics. After the mapping is completed, the adjustment parameters used to adjust the vehicle steering system response characteristics are output. These adjustment parameters are quantitative parameters that directly correspond to the adjustment requirements of the steering system's feedback rules and execution response methods.

[0066] For example, the fused decision features output by S122 are input into the feature validity verification model. The validity verification model verifies the feature's dimensionality completeness (ensuring it is a preset 5 dimensions) and anomaly confidence validity (ensuring the anomaly confidence is greater than a preset threshold of 0.5) according to preset rules. If the verification passes, the valid fused decision features are output and proceed to the subsequent mapping stage. If the verification fails (e.g., feature dimensions are missing, or anomaly confidence is insufficient), the process returns to S122 to re-fuse the features, ensuring the features input to the mapping model are valid. The valid fused decision features that pass the verification are then input into a pre-trained feature-parameter mapping model (this feature-parameter mapping model is trained based on a large number of driver direction correction anomaly scenario samples and vehicle steering control parameter samples, and has undergone iterative optimization). The feature-parameter mapping... The feature-parameter mapping model automatically maps abstract anomaly representations to the vehicle's lateral control parameter space by analyzing the anomaly type (high-frequency sudden anomaly) and anomaly level (moderate) in the fused decision features, realizing the transformation from "anomaly attribute to control parameter". During the mapping process, the logic of "positive correlation between anomaly level and adjustment range" is followed to ensure that the adjustment parameters can accurately match the current abnormal behavior. The feature-parameter mapping model outputs adjustment parameters for adjusting the response characteristics of the vehicle's steering system. These adjustment parameters are quantified control parameters that are directly related to the steering system's response characteristics, specifically including: steering feedback damping coefficient adjustment parameters, steering assist gain coefficient adjustment parameters, and steering response hysteresis time adjustment parameters. The adjustment direction and magnitude of the adjustment parameters match the "moderate high-frequency sudden direction correction anomaly" represented by the fused decision features.

[0067] In this embodiment, by mapping the fusion decision features that characterize the abnormal attributes of driver direction correction to the vehicle lateral control parameter space, the abstract abnormal information is transformed into specific adjustment parameters that can be directly implemented to adjust the response characteristics of the steering system, thus realizing the connection between the "abnormal judgment result" and the "control execution action". At the same time, the mapping process strictly matches the control logic of the steering system to ensure that the adjustment parameters can correspond to the type and level of abnormal behavior, providing a quantitative and executable core basis for S132 to generate targeted control strategies.

[0068] S132. Generate a vehicle control strategy based on the adjusted parameters.

[0069] Specifically, the system acquires vehicle driving condition data, which reflects the vehicle's current driving environment and operating status. This data is the core basis for adjusting and correcting parameters, ensuring that the corrected parameters are suitable for the current driving scenario. Based on the acquired real-time driving condition data, the system uses a preset condition adaptation model to specifically correct the adjustment parameters output by S131. The core logic of this correction is "matching the operating condition characteristics with the parameter adjustment range," that is, adjusting the specific values ​​of the parameters according to the steering system control requirements under different operating conditions to obtain target control parameters suitable for the current operating conditions. Based on the corrected target control parameters and combined with the vehicle steering system control logic, a complete vehicle control strategy is generated. The core content of this vehicle control strategy is to adjust the steering system's feedback rules and execution response methods to the driver's steering wheel operations.

[0070] For example, the vehicle's driving condition data is collected in real time through the vehicle's onboard operating condition data acquisition module, specifically including: real-time vehicle speed data, road surface adhesion coefficient data, and road condition data. After preprocessing, the collected data is input into the operating condition adaptation model as the core basis for adjusting parameters. The adjustment parameters output by S131 and the collected real-time driving condition data are simultaneously input into the pre-trained operating condition adaptation model (this model is trained based on samples with different driving conditions and different anomaly levels, enabling adaptation between operating conditions and control parameters). The operating condition adaptation model analyzes the current operating condition characteristics (urban expressway, moderate speed, normal road surface adhesion coefficient), and, combined with the "moderate high-frequency sudden direction correction anomaly" represented by the fusion decision feature, makes targeted adjustments to the adjustment parameters: for example, for urban expressway conditions, the adjustment range of steering response lag time is appropriately reduced to avoid affecting vehicle handling. The system maintains flexibility while retaining the core adjustment logic for steering feedback damping and power assist gain. After correction, it outputs target control parameters adapted to the current operating conditions, ensuring that the target control parameters are both suitable for abnormal behavior and fit the current driving scenario. The corrected target control parameters are then input into the control strategy generation model. The control strategy generation model, combined with the control logic of the vehicle steering system, transforms the target control parameters into directly executable control commands, generating a complete vehicle control strategy. This includes: adjusting the steering system feedback rules (increasing steering wheel torque feedback damping to guide the driver to reduce high-frequency abnormal correction actions) and adjusting the steering system execution response mode (appropriately reducing steering assist gain, optimizing steering response rate, and avoiding oversteering). The generated control strategy format matches the control interface of the vehicle steering system, allowing it to be directly recognized and executed by the steering system, achieving precise and adaptive intervention for abnormal driver steering correction behavior.

[0071] In this embodiment, by introducing real-time vehicle driving condition data, the adjustment parameters output by S131 are specifically corrected, avoiding the problems of over-intervention or under-intervention that occur when fixed adjustment parameters are used under different vehicle speeds and road conditions. The resulting vehicle control strategy can both match the abnormal behavior of directional correction represented by the fusion decision feature and adapt to the current driving conditions of the vehicle, realizing dual control of "abnormal behavior adaptation + driving condition adaptation". This effectively suppresses the driving safety risks caused by abnormal correction behavior and maximizes the driving experience and vehicle handling smoothness, further improving the practicality and reliability of the entire vehicle control scheme.

[0072] In the above scheme, steering system adjustment parameters are obtained by mapping the fused decision features to the vehicle's lateral control parameter space. The vehicle control strategy is generated after the adjustment parameters are corrected by combining the vehicle's real-time driving condition data. This can transform abstract abnormal characterization information into directly executable quantitative control parameters. At the same time, it makes the control strategy compatible with the vehicle's current driving condition, avoiding problems of over-intervention or under-intervention. It achieves adaptive intervention for abnormal behavior of driver's directional correction, effectively improving the vehicle's lateral driving safety while taking into account the driver's driving experience and vehicle handling smoothness.

[0073] In some embodiments, such as Figure 4 As shown, vehicle control methods based on direction correction anomalies also include: S14. After the vehicle control strategy is executed, obtain feedback data on the vehicle status and driver operation.

[0074] Specifically, after the vehicle control strategy generated by S132 is executed by the vehicle steering system, the feedback data acquisition process is immediately triggered to ensure that the acquired data can truly reflect the execution effect of the control strategy and avoid feedback distortion caused by time delay. Through the preset data acquisition module on the vehicle, two types of core feedback data are acquired simultaneously: vehicle status data (reflecting the vehicle's operational stability after control) and driver operation data (reflecting the driver's response and adjustment to control intervention). The acquired raw feedback data is preprocessed to eliminate data noise, timing misalignment, outliers, and other issues, ensuring the authenticity and standardization of the feedback data. The preprocessed effective feedback data is then output.

[0075] In this embodiment, by collecting feedback data on vehicle status and driver operation in real time after the vehicle control strategy is executed, the actual execution effect of the control strategy can be captured (such as whether abnormal behavior is effectively suppressed and whether steering intervention is appropriate). This feedback data, together with multi-source operating data, fused decision features, and control parameters, provides real-world sample support for the optimization of S15, ensuring that subsequent optimizations can fit the actual driving scenario, avoiding optimizations from deviating from actual needs, and laying the foundation for the continuous iteration of the entire technical solution.

[0076] S15. Based on feedback data, optimize the extraction rules and fusion weights corresponding to time-series frequency features, multi-dimensional driving behavior features, and supplementary driving behavior features.

[0077] Specifically, feedback data is integrated with multi-source operational data to obtain optimized driving behavior samples, which contain complete link information of "original behavior - control intervention - feedback result". Based on the optimized driving behavior samples, extraction rules are used to determine the feature values ​​corresponding to time-series frequency features, multi-dimensional driving behavior features, and supplementary driving behavior features, which serve as the basis for deviation analysis. The deviation between feature values ​​and sample values ​​is determined, which reflects the insufficient representation accuracy of the current extraction rules. With the goal of minimizing the deviation, the extraction rules of the three types of features (such as feature extraction threshold, time-series window, and semantic encoding logic) are adjusted through a preset optimization algorithm to improve the feature representation accuracy. With the improvement of the confidence of the fused decision features as the optimization goal, the contribution of the three types of features is recalculated based on the optimized driving behavior samples, and the fusion weights are updated to ensure that the weights match the actual representation ability and control effect of the features, thus completing the entire optimization process.

[0078] For example, the effective feedback data from S14 and the multi-source operational data from S11 are input into the sample integration model. The sample integration model, aligned with millisecond-level timestamps, integrates the multi-source operational data before control (e.g., steering wheel angle and steering torque before control), the control parameters during control (target control parameters of S132), and the feedback data after control (e.g., yaw rate and driver operation adjustments after control), generating an optimized driving behavior sample containing the complete "before control - during control - after control" chain. This optimized driving behavior sample is then input into the existing feature extraction model (using the currently unoptimized extraction rules) to extract time-series frequency features, multi-dimensional driving behavior features, and feature values ​​corresponding to supplementary driving behavior features, yielding the current representation results for the three types of features. Through a deviation calculation model, the extracted feature values ​​are compared with the corresponding sample values ​​(real behavior representation values) in the optimized driving behavior sample. The mean squared error algorithm is used to calculate the deviation between the two, identifying the key links in the current extraction rules that cause representation deviations. (For example, unreasonable temporal window settings or semantic encoding threshold deviations) provide a clear direction for extraction rule optimization. The deviation calculation results are input into the extraction rule optimization model, which takes minimizing the deviation as its core optimization objective and uses the gradient descent algorithm to iteratively optimize the extraction rules for the three types of features. For example, for temporal frequency features, the size of the temporal window for extracting high-frequency fluctuations is adjusted to improve the accuracy of capturing high-frequency sudden anomalies; for supplementary driving behavior features, the semantic encoding threshold is optimized to improve the accuracy of discriminating atypical anomalies. After optimization, the updated feature extraction rules are output to replace the original extraction rules in S11, thereby improving the feature extraction accuracy. The optimized driving behavior samples are input into the weight update model, which takes improving the confidence of the fused decision features as its optimization objective. It recalculates the temporal response consistency index and behavioral semantic consistency index of the three types of features and updates the contribution of each type of feature. Based on the updated contribution, the fusion weights of the three types of features are re-determined to obtain the optimized dynamic weights.

[0079] In this embodiment, by integrating feedback data with original multi-source operational data, optimized samples with real control effect verification are obtained, providing reliable supervision for optimization. Optimizing extraction rules with the goal of minimizing deviation can improve the representation accuracy of the three types of features for directional correction behavior and reduce anomaly identification bias. Updating fusion weights with the goal of improving the confidence of fusion decision features can continuously optimize the feature fusion effect, allowing the entire anomaly identification and fusion system to continuously adapt to changes in driver driving habits and different driving scenarios. Ultimately, this achieves a continuous improvement in anomaly identification accuracy and control intervention effectiveness, further reducing the probability of false detection and missed detection of abnormal behavior, improving vehicle lateral driving safety and driving experience, and ensuring the technical solution has long-term adaptability and practicality.

[0080] In the above scheme, feedback data on vehicle status and driver operation are obtained after the vehicle control strategy is executed. Based on the feedback data, optimized driving behavior samples are constructed. The feature extraction rules are optimized with the goal of minimizing the deviation between feature values ​​and sample values. At the same time, the optimized driving behavior samples are input into the weight update model. With the goal of improving the confidence of fusion decision features, the double consistency index is recalculated, and the feature contribution and fusion weights are updated. This can construct a complete closed loop of "control execution - effect feedback - iterative optimization", continuously improve the accuracy of feature extraction and the anomaly representation accuracy of fusion decision features, and enable the overall recognition and control system to adaptively match different driving scenarios and driver driving habits, further improving the accuracy of direction correction anomaly recognition and the effectiveness of vehicle control intervention.

[0081] In addition, such as Figure 5 As shown, Figure 5 This is a schematic diagram of a vehicle control device based on a direction correction anomaly, provided in an embodiment of this application. The vehicle control device 500 based on the direction correction anomaly includes: The determination module 501 is used to determine the vehicle's temporal frequency characteristics, multi-dimensional driving behavior characteristics, and supplementary driving behavior characteristics based on the vehicle's multi-source operating data. The temporal frequency characteristics are used to characterize the temporal characteristics of the driver's high-frequency and sudden directional correction behaviors, the multi-dimensional driving behavior characteristics are used to characterize the directional correction behavior characteristics, and the supplementary driving behavior characteristics are used to characterize the discrimination characteristics of abnormal directional correction behaviors that are not predefined. The fusion module 502 is used to fuse time-series frequency features, multi-dimensional driving behavior features and supplementary driving behavior features to obtain fused decision features, which are used to characterize abnormal driver direction correction behavior. The generation module 503 is used to generate a vehicle control strategy based on the fused decision features. The vehicle control strategy is used to correct abnormal behavior in response to driver direction and adjust the feedback rules and execution response mode of the steering system to the driver's steering wheel operation.

[0082] The above scheme uses multi-source vehicle operation data as the basis for identification. Compared with single-dimensional steering data, it provides more comprehensive and multi-dimensional data source support for identifying driver directional correction behavior. Based on this, three types of complementary and clearly defined features are designed and extracted. The combination of these three types of features achieves full-scene and full-dimensional coverage of typical and atypical, transient and steady-state abnormal directional correction behaviors of drivers, improving the completeness and accuracy of abnormal behavior capture from the source. Furthermore, by fusing the three complementary features, a fused decision feature is obtained. Compared with identification results relying on a single feature, the cross-validation of multi-dimensional features effectively avoids the susceptibility of single features to data noise interference. By eliminating the large deviation, the probability of false positives and false negatives in abnormal behavior is significantly reduced, providing a high-confidence and highly focused decision-making basis for the generation of subsequent control strategies. Finally, based on the fused decision features, a targeted vehicle control strategy is generated. For the detected abnormal behavior of driver direction correction, the feedback rules and execution response mode of the steering system to the driver's steering wheel operation are directly adjusted, realizing a direct closed loop from abnormal behavior identification results to steering control execution actions. This allows the intervention action of the steering system to match the attributes of the currently detected abnormal behavior. Through the targeted adjustment of steering feedback rules and execution response mode, the driving safety risks such as vehicle yaw instability and lane departure caused by the driver's unintentional abnormal direction correction behavior can be effectively suppressed.

[0083] In one specific embodiment, the fusion module 502 is further configured to: Determine the weights corresponding to time-series frequency features, multi-dimensional driving behavior features, and supplementary driving behavior features; Based on weights, time-series frequency features, multi-dimensional driving behavior features, and supplementary driving behavior features are fused to obtain fused decision features.

[0084] In one specific embodiment, the fusion module is further configured to: After aligning the temporal frequency features, multi-dimensional driving behavior features, and supplementary driving behavior features to the same time base and feature dimension, the temporal change trend matching degree of each pair of features within the same time window for the driver's directional correction actions is determined, and the temporal response consistency index is obtained. Extract temporal frequency features, temporal multi-dimensional driving behavior features, and direction correction behavior representation semantic vectors corresponding to supplementary driving behavior features respectively, determine the semantic vector space matching degree of each pair of features, and obtain the behavior semantic consistency index. Based on the temporal response consistency index and the behavioral semantic consistency index, the weights corresponding to temporal frequency features, multi-dimensional driving behavior features and supplementary driving behavior features are determined.

[0085] In one specific embodiment, the fusion module 502 is further configured to: Based on the temporal response consistency index and the behavioral semantic consistency index, the confidence scores corresponding to temporal frequency features, multi-dimensional driving behavior features and supplementary driving behavior features are determined respectively. The confidence score was normalized, and the contributions of time-series frequency features, multi-dimensional driving behavior features and supplementary driving behavior features to the identification of abnormal driver direction correction behavior were determined. Based on contribution, the weights corresponding to time-series frequency features, multi-dimensional driving behavior features, and supplementary driving behavior features are determined.

[0086] In one specific embodiment, the generation module 503 is further configured to: The fused decision features are mapped to the vehicle lateral control parameter space to obtain adjustment parameters for adjusting the response characteristics of the vehicle steering system; Based on the adjusted parameters, a vehicle control strategy is generated.

[0087] In one specific embodiment, the generation module 503 is further configured to: Obtain vehicle operating condition data; The adjustment parameters are corrected based on driving condition data to obtain the target control parameters; A vehicle control strategy is generated based on the target control parameters.

[0088] In one specific embodiment, the vehicle control device 500 based on direction correction anomalies further includes an acquisition module: The acquisition module is used to acquire feedback data on vehicle status and driver operation after the vehicle control strategy is executed. Based on feedback data, the extraction rules and fusion weights corresponding to time-series frequency features, multi-dimensional driving behavior features, and supplementary driving behavior features are optimized.

[0089] In one specific embodiment, the acquisition module is further configured to: By integrating feedback data with multi-source operational data, optimized driving behavior samples are obtained. Based on optimized driving behavior samples, extraction rules are used to determine the feature values ​​corresponding to time-series frequency features, multi-dimensional driving behavior features, and supplementary driving behavior features; Determine the deviation between the feature values ​​and the sample values; Optimize the extraction rules with the goal of minimizing the bias; With the goal of improving the confidence of fused decision features, the fusion weights are updated based on optimized driving behavior samples.

[0090] In one specific embodiment, the determining module 501 is further configured to: After preprocessing the multi-source operation data, the temporal variation pattern of the driver's direction correction actions, direction correction behavior, and the discrimination characteristics of predefined abnormal direction correction behavior are extracted from the multi-source operation data through a preset feature extraction strategy. Based on the temporal variation patterns of driver directional correction actions, the discrimination characteristics of directional correction behavior and predefined abnormal directional correction behavior, temporal frequency features, multi-dimensional driving behavior features and supplementary driving behavior features are determined respectively.

[0091] Regarding the apparatus in the above embodiments, the specific manner in which each unit performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0092] This embodiment also provides a vehicle, including: the vehicle executing any of the optional vehicle control methods based on direction correction anomalies described above, thus achieving the same effect as the above implementation methods.

[0093] The beneficial effects of the above embodiments can be referred to the beneficial effects of the corresponding methods provided above, and will not be repeated here.

[0094] Through the above description of the embodiments, those skilled in the art will understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0095] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0096] In the description of this application, it should be understood that if the terms "upper", "lower", "front", "rear", "left" and "right" are used to indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the position or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application.

[0097] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0098] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A vehicle control method based on direction correction anomalies, characterized in that, The method includes: Based on the vehicle's multi-source operation data, the vehicle's temporal frequency characteristics, multi-dimensional driving behavior characteristics, and supplementary driving behavior characteristics are determined. The temporal frequency characteristics are used to characterize the temporal characteristics of the driver's high-frequency and sudden directional correction behavior. The multi-dimensional driving behavior characteristics are used to characterize the directional correction behavior characteristics. The supplementary driving behavior characteristics are used to characterize the discrimination characteristics of abnormal directional correction behavior that is not predefined. The time-series frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features are fused to obtain fused decision features, which are used to characterize abnormal driver direction correction behavior. Based on the fused decision features, a vehicle control strategy is generated. The vehicle control strategy is used to correct abnormal behavior in response to driver direction and adjust the feedback rules and execution response mode of the steering system to the driver's steering wheel operation.

2. The vehicle control method based on direction correction anomaly according to claim 1, characterized in that, The process of fusing the temporal frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features to obtain fused decision features includes: Determine the weights corresponding to the time-series frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features; The time-series frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features are fused based on the weights to obtain the fused decision features.

3. The vehicle control method based on direction correction anomaly according to claim 2, characterized in that, Determining the weights corresponding to the time-series frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features includes: After aligning the temporal frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features to the same time reference and feature dimension, the temporal change trend matching degree of each pair of features within the same temporal window is determined to obtain the temporal response consistency index. Extract the direction correction behavior representation semantic vectors corresponding to the temporal frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features respectively, determine the semantic vector space matching degree of each pair of features, and obtain the behavior semantic consistency index. Based on the temporal response consistency index and the behavioral semantic consistency index, the weights corresponding to the temporal frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features are determined.

4. The vehicle control method based on direction correction anomaly according to claim 3, characterized in that, The step of determining the weights corresponding to the temporal frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features based on the temporal response consistency index and the behavioral semantic consistency index includes: Based on the temporal response consistency index and the behavioral semantic consistency index, the confidence scores corresponding to the temporal frequency features, the multi-dimensional driving behavior features and the supplementary driving behavior features are determined respectively. The confidence score is normalized to determine the contribution of the time-series frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features to the driver's direction correction behavior in the process of identifying abnormal driver direction correction behavior. Based on the contribution, the weights corresponding to the time-series frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features are determined.

5. The vehicle control method based on direction correction anomaly according to claim 1, characterized in that, The process of generating a vehicle control strategy based on the fused decision features includes: The fused decision features are mapped to the vehicle lateral control parameter space to obtain adjustment parameters for adjusting the response characteristics of the vehicle steering system; The vehicle control strategy is generated based on the adjusted parameters.

6. The vehicle control method based on direction correction anomaly according to claim 5, characterized in that, The process of generating the vehicle control strategy based on the adjusted parameters includes: Obtain vehicle operating condition data; The adjustment parameters are corrected based on the driving condition data to obtain the target control parameters; The vehicle control strategy is generated based on the target control parameters.

7. The vehicle control method based on direction correction anomaly according to claim 1, characterized in that, The method further includes: After the vehicle control strategy is executed, feedback data on vehicle status and driver operation are acquired. Based on the feedback data, the extraction rules and fusion weights corresponding to the time-series frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features are optimized.

8. The vehicle control method based on direction correction anomaly according to claim 7, characterized in that, The step of optimizing the extraction rules and fusion weights corresponding to the time-series frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features based on the feedback data includes: The feedback data is integrated with the multi-source operational data to obtain an optimized driving behavior sample; Based on the optimized driving behavior sample, the feature values ​​corresponding to the time-series frequency feature, the multi-dimensional driving behavior feature, and the supplementary driving behavior feature are determined using the extraction rules. Determine the deviation between the feature value and the sample value; The extraction rules are optimized with the goal of minimizing the aforementioned deviation. With the improvement of the confidence level of the fused decision features as the optimization objective, the fusion weights are updated based on the optimized driving behavior samples.

9. The vehicle control method based on direction correction anomaly according to claim 1, characterized in that, The determination of the vehicle's temporal frequency characteristics, multi-dimensional driving behavior characteristics, and supplementary driving behavior characteristics based on the vehicle's multi-source operational data includes: After preprocessing the multi-source operation data, the temporal variation pattern of the driver's direction correction actions, direction correction behavior, and the discrimination characteristics of predefined abnormal direction correction behavior are extracted from the multi-source operation data through a preset feature extraction strategy. Based on the temporal variation pattern of the driver's directional correction actions, the directional correction behavior, and the discrimination characteristics of predefined abnormal directional correction behaviors, the temporal frequency features, the multi-dimensional driving behavior features, and the supplementary driving behavior features are determined respectively.

10. A vehicle, characterized in that, include: The vehicle performs the vehicle control method based on direction correction anomalies as described in any one of claims 1 to 9.

Images