Gradient correction and training method and device of gradient correction model, and storage medium

By fusing vehicle operating status data with the baseline slope using a self-attention mechanism, the problem of inaccurate slope calculation caused by strong sensor dependence is solved, thus achieving accurate slope prediction and stable vehicle control.

CN122432497APending Publication Date: 2026-07-21YUANYI HUANYU (SHANGHAI) TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
YUANYI HUANYU (SHANGHAI) TECHNOLOGY CO LTD
Filing Date
2026-03-04
Publication Date
2026-07-21

AI Technical Summary

Technical Problem

In existing technologies, road slope estimation methods based on vehicle sensors are easily affected by external environmental interference, resulting in insufficient accuracy in slope calculation and failing to meet the precise requirements of intelligent vehicle control.

Method used

A self-attention mechanism is adopted to fuse vehicle operating status data with the baseline slope difference. By constructing input features and attention weights of historical time steps, the slope correction model is trained and predicted. The self-attention mechanism is used to calculate attention weights and fuse slope difference data to improve the accuracy of slope estimation.

Benefits of technology

It achieves precise correction of the benchmark slope, improves the accuracy of slope prediction, and ensures stable operation and control of vehicles under different slope conditions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122432497A_ABST
    Figure CN122432497A_ABST
Patent Text Reader

Abstract

The embodiment of the application discloses a slope correction method and a slope correction model training method, device and storage medium. The specific implementation scheme is: based on the vehicle running state data and the reference slope of a plurality of historical time steps before the current time step, the input features of each historical time step are constructed, the attention weights of the input features of each historical time step are calculated based on the self-attention mechanism, the attention weight of the input features of the first historical time step to the input features of the first historical time step is fused with the difference data of the reference slope between the first historical time step and the first historical time step, the fused weight is obtained, the input features of each historical time step are fused based on the fused weight, and the fused features are obtained, so as to determine the corrected slope of the current time step based on the fused features. Based on the scheme, the reference slope can be accurately corrected, so as to improve the accuracy of the predicted slope.
Need to check novelty before this filing date? Find Prior Art