CN-121973848-A - Soft dead point rebound suppression method for steer-by-wire system

CN121973848ACN 121973848 ACN121973848 ACN 121973848ACN-121973848-A

Abstract

The invention provides a soft dead center rebound suppression method for a steer-by-wire system, which relates to the technical field of steer-by-wire, and comprises the steps of acquiring vehicle multidimensional data corresponding to a vehicle in real time, wherein the vehicle multidimensional data comprise steering-by-wire basic data, driver hand pressure distribution signals and driver sight focal signals, judging and identifying a high-risk oversteer rebound scene of the vehicle based on the steering-by-wire basic data, and extracting multidimensional features based on the steering-by-wire basic data, the driver hand pressure distribution signals and the driver sight focal signals synchronously acquired in the high-risk oversteer rebound scene to obtain a multidimensional feature set, identifying the steering intention of the driver based on a steering intention identification model which completes training in advance based on the multidimensional feature set, and matching a corresponding differential rebound suppression strategy from a database based on the steering intention of the driver.

Inventors

YANG HAOWEI
ZHANG RUILONG

Assignees

杭州湘滨电子科技股份有限公司

Dates

Publication Date: 20260505
Application Date: 20260309

Claims (8)

1. A soft dead center rebound suppression method for a steer-by-wire system, the method comprising: The method comprises the steps of acquiring vehicle multidimensional data corresponding to a vehicle in real time, wherein the vehicle multidimensional data comprise steering-by-wire basic data, a driver hand pressure distribution signal and a driver sight focus signal; Judging and identifying the high-risk over-bending rebound scene of the vehicle based on the high-risk over-bending rebound scene judging rule and the wire control steering basic data; When the vehicle is in a high-risk oversteering rebound scene, carrying out multi-dimensional feature extraction based on the wire control steering basic data, the driver hand pressure distribution signals and the driver sight focus signals which are synchronously acquired in the high-risk oversteering rebound scene, so as to obtain a multi-dimensional feature set; Based on dimension characteristics in the multi-dimension characteristic set, based on a steering intention identification model which completes training in advance, identifying the steering intention of a driver, wherein the steering intention of the driver comprises active loosening hands, fatigue shake and stable disc fine adjustment; based on the driver steering intent, a corresponding differential rebound suppression strategy is matched from the database.
2. The method for suppressing rebound of a soft dead point for a steer-by-wire system according to claim 1, wherein the judging and identifying the high-risk over-bend rebound scene of the vehicle based on the high-risk over-bend rebound scene judging rule and the steer-by-wire basic data comprises: extracting precursor features of multiple dimensions based on a hand feeling simulator corner signal, a hand feeling simulator rotating speed signal, a steering wheel hand moment signal and a steering wheel soft dead point angle in the steering-by-wire basic data; Carrying out normalization processing on each precursor feature to obtain a multi-dimensional standard precursor feature set; based on a multidimensional standard precursor feature set and based on a preset feature concurrency triggering rule strongly related to the high-risk overstretched rebound scene, judging whether to trigger a concurrency event; for each triggered concurrent event, performing nonlinear confidence strengthening evaluation based on the multidimensional standard precursor feature set, and outputting a comprehensive confidence score; comparing the output comprehensive confidence score with a preset scene discrimination threshold; and if the comprehensive confidence score is higher than the scene discrimination threshold, judging that the vehicle is currently in the high-risk over-bending rebound scene.
3. The method of soft dead center rebound suppression for a steer-by-wire system of claim 2, wherein said performing a non-linear confidence reinforcement evaluation based on a multi-dimensional standard precursor feature set for each triggered concurrent event, outputting a composite confidence score comprises: Based on a multi-dimensional standard precursor feature set and on a single variable statistical control limit corresponding to each precursor feature, calculating a real-time value of the precursor feature and a first amplitude of the preset single variable control limit, and further obtaining a reference contribution degree of each precursor feature; based on the multidimensional standard precursor feature set, corresponding real-time multivariable is calculated Statistics, calculate real-time multivariable Statistics and preset multivariable Counting a second amplitude of the control limit, and further calculating to obtain interaction contribution degree; Based on the interaction contribution degree, calculating to obtain a dynamic overflow gain value of the reference contribution degree of each precursor feature by a preset nonlinear gain function; based on the dynamic overflow gain value, enhancing the reference contribution degree of the corresponding precursor features to obtain the final contribution degree of each precursor feature after the coupling effect correction; based on the final contribution of each precursor feature, fusion is performed to form a comprehensive confidence score.
4. The method for suppressing soft dead center rebound of a steer-by-wire system according to claim 3, wherein said obtaining a multi-dimensional feature set while performing multi-dimensional feature extraction based on said steer-by-wire basic data, a driver hand pressure distribution signal, and a driver line of sight focus signal synchronously acquired in said high-risk oversteer rebound scene comprises: When the vehicle is in a high-risk oversteering rebound scene, preprocessing the synchronously acquired steering-by-wire basic data, a driver hand pressure distribution signal and a driver sight focus signal to obtain a corresponding derivative signal set, wherein the derivative signal set comprises a steering wheel hand moment signal, a steering wheel corner speed signal, a vehicle body transverse acceleration signal, a hand pressure sum signal, a pressure center track signal and a driver sight focus signal; Based on the derived signal set, performing multidimensional feature extraction to obtain a first extracted feature, a second extracted feature, a third extracted feature and a fourth extracted feature; and combining the extracted first extraction feature, second extraction feature, third extraction feature and fourth extraction feature to form the multi-dimensional feature set.
5. The method of claim 4, wherein the scene discrimination threshold is dynamically adjusted based on a threshold dynamic adjustment strategy, the threshold dynamic adjustment strategy being: Recording comprehensive confidence scores of each trigger high-risk overstock scene judgment according to a time sequence in a sliding time window with a preset length to form a corresponding historical scoring sequence; At the current time, a historical scoring sequence of the last time corresponding to the current time in a database is called, wherein the historical scoring sequence is a set of comprehensive confidence scores for triggering high-risk overstock scene judgment each time in a sliding time window with a preset length according to a time sequence; periodically calculating key statistical features corresponding to the historical scoring sequence based on the historical scoring sequence of the last moment corresponding to the current moment, wherein the key statistical features are respectively a concentration index and a trend intensity index; When the concentration index is larger than a preset concentration threshold and the trend intensity index is larger than zero, the scene discrimination threshold is adjusted downwards based on the first function; when the concentration index is smaller than a preset concentration threshold and the trend intensity index is continuously smaller than zero, the scene discrimination threshold is adjusted upwards based on a second function; And in other cases, the scene discrimination threshold value at the previous moment is kept unchanged and is used as the scene discrimination threshold value at the current moment.
6. The method of soft dead center rebound suppression for a steer-by-wire system of claim 5, wherein matching the corresponding differential rebound suppression strategy from the database based on driver steering intent comprises: Selecting all candidate inhibition strategies related to the steering intention of a driver from a database based on the steering intention of the driver currently corresponding to the driver to form a candidate strategy set, wherein a plurality of candidate inhibition strategies related to the steering intention of each driver are stored in the database, and each candidate inhibition strategy is related to a dynamic efficiency score; The dynamic efficacy score is updated based on the effect of the candidate suppression strategy history after being executed; Correcting the dynamic performance scores in the strategy set to be selected based on the comprehensive confidence scores output by the judgment; And selecting the candidate suppression strategy with the highest dynamic efficiency score as the differentiated rebound suppression strategy to be currently executed based on each corrected candidate suppression strategy.
7. The method for soft dead center rebound suppression of a steer-by-wire system of claim 6, wherein the training process of the pre-trained steering intent discrimination model is: The method comprises the steps of retrieving a historical multi-dimensional feature set from a database, and marking steering intention of the historical multi-dimensional feature set; Constructing a turning intention identification model, and dividing the hyper-parameters corresponding to the turning intention identification model into three hyper-parameter heterogeneous subsets based on a model network structure of the turning intention identification model; based on the three screened super-parameter heterogeneous subsets, generating a plurality of super-parameter combinations by Latin hypercube sampling; taking the historical multi-dimensional feature set as input, inputting the historical multi-dimensional feature set into a steering intention identification model corresponding to each super-parameter combination, and outputting corresponding steering intention result data; based on steering intention result data and corresponding labeling content, evaluating and calculating comprehensive identification performance corresponding to each super-parameter set, and screening super-parameter combinations arranged in the previous k into a candidate set; And performing differential variation on each super-parameter combination in the candidate set to obtain a new super-parameter combination, re-inputting the historical multi-dimensional feature set until the maximum iteration number is reached, and selecting the super-parameter combination with the maximum comprehensive identification performance as a final super-parameter set of the steering intention identification model, thereby completing training of the steering intention identification model.
8. The method of soft dead center rebound suppression for a steer-by-wire system of claim 7, wherein combining the extracted first, second, third and fourth extracted features to form the multi-dimensional feature set comprises: pairing every two based on the first extraction feature, the second extraction feature, the third extraction feature and the fourth extraction feature to form extraction feature pairs, and calculating a bidirectional association strength value between each extraction feature pair; Based on the two-way association strength value, carrying out association grade classification on each extracted feature pair based on a preset first strength threshold value and a preset second strength threshold value, wherein the association grade comprises a strong association feature pair, a medium association feature pair and a weak association feature pair; and based on the corresponding association level of each extracted feature pair, a corresponding feature processing strategy is called to form each new extracted feature, and after normalization processing is carried out on each new extracted feature, the multi-dimensional feature sets are formed by combining.

Description

Soft dead point rebound suppression method for steer-by-wire system Technical Field The invention relates to the technical field of steer-by-wire, in particular to a soft dead center rebound suppression method for a steer-by-wire system. Background The drive-by-wire steering system realizes dynamic adjustment of steering transmission ratio through mechanical decoupling, which is a key technology for automobile intellectualization, but the soft dead point rebound suppression under a high dynamic scene still has a technical bottleneck. Under the scenes of high-speed overbending and the like, the wire control steering mechanism is easy to reach a soft dead point and generates rebound, so that the steering wheel is not expected to reversely rotate, the operation of a driver is disturbed, the driving risk is increased, and the soft dead point rebound inhibition is a core technology pain point. The existing scheme mostly makes a suppression strategy based on single-dimensional data such as steering angle, vehicle speed and the like, and does not consider the complexity of steering intention of a driver in a high dynamic scene. When the vehicle is bent at a high speed, the driver actively loosens hands, tires and shakes, the operation characteristics of the intention such as steady disc fine adjustment are overlapped, and the real steering requirement is difficult to accurately distinguish by the single-dimensional data. The intention is to resolve the misalignment and easily cause insufficient suppression or excessive interference, system delay and signal noise under the high dynamic scene are overlapped, the resolution difficulty is further increased, and the existing scheme cannot meet the safety operation requirement. Therefore, the soft dead point rebound suppression method for the steer-by-wire system can realize accurate identification of steering intention of a driver in a high-risk over-bending rebound scene of the vehicle, and greatly improve safety and reliability of the vehicle in the high-risk over-bending rebound scene. Disclosure of Invention In order to solve the technical problems, the invention aims to provide a soft dead point rebound suppression method for a steer-by-wire system, which can realize accurate identification of steering intention of a driver in a high-risk over-bending rebound scene of a vehicle and greatly improve safety and reliability of the vehicle in the high-risk over-bending rebound scene. In order to achieve the above purpose, the invention provides a soft dead center rebound suppression method for a steer-by-wire system, comprising the following steps: The method comprises the steps of acquiring vehicle multidimensional data corresponding to a vehicle in real time, wherein the vehicle multidimensional data comprise steering-by-wire basic data, a driver hand pressure distribution signal and a driver sight focus signal; Judging and identifying the high-risk over-bending rebound scene of the vehicle based on the high-risk over-bending rebound scene judging rule and the wire control steering basic data; When the vehicle is in a high-risk oversteering rebound scene, carrying out multi-dimensional feature extraction based on the wire control steering basic data, the driver hand pressure distribution signals and the driver sight focus signals which are synchronously acquired in the high-risk oversteering rebound scene, so as to obtain a multi-dimensional feature set; Based on dimension characteristics in the multi-dimension characteristic set, based on a steering intention identification model which completes training in advance, identifying the steering intention of a driver, wherein the steering intention of the driver comprises active loosening hands, fatigue shake and stable disc fine adjustment; based on the driver steering intent, a corresponding differential rebound suppression strategy is matched from the database. Preferably, the judging and identifying the high-risk over-bending rebound scene of the vehicle based on the high-risk over-bending rebound scene judging rule and the steering-by-wire basic data comprises the following steps: extracting precursor features of multiple dimensions based on a hand feeling simulator corner signal, a hand feeling simulator rotating speed signal, a steering wheel hand moment signal and a steering wheel soft dead point angle in the steering-by-wire basic data; Carrying out normalization processing on each precursor feature to obtain a multi-dimensional standard precursor feature set; based on a multidimensional standard precursor feature set and based on a preset feature concurrency triggering rule strongly related to the high-risk overstretched rebound scene, judging whether to trigger a concurrency event; for each triggered concurrent event, performing nonlinear confidence strengthening evaluation based on the multidimensional standard precursor feature set, and outputting a comprehensive confidence score; comparing the output compre