TRB Pubsindex
Text Size:
x

« Prev Next »

Return to Results | New Search

Bookmark and Share

Title:

A Method for Short-Term Prediction of Driving Risk Based on the Light Gradient Boosting Model

Accession Number:

01872486

Record Type:

Component

Abstract:

With the gradual popularization of advanced driver assistance systems, a large amount of refined information such as the vehicle itself and its surrounding environment can be obtained during driving, providing data support for short-term driving risk prediction. This study proposes a threshold update method for risk indicators based on Monte Carlo simulation to further extract risk events; the optimal risk prediction advance time and window width are determined by a variable sliding window method; A short-term risk prediction method integrating combines sample imbalance processing, feature screening, model prediction and risk influencing factor analysis. The method is validated and tested on a dataset of 88 real car driving. The test results show that the optimal risk prediction window advance time and window length are 1.6 seconds and 1.2 seconds respectively; LGBM model has good prediction effect and predicted F1 scores reaches 87.90; the prediction effect of using different data sources is different, and the data containing all the information of the pedestrian, vehicle and surroundings has the best prediction effect. In addition, the SHAP analysis show that speed, TTCi and accelerator are particularly important for perception of driving risks. The method proposed in this paper can well meet the risk prediction requirements of future ADAS. Through various types of vehicle perception data collected in real time, it can identify and predict driving risks in advance in a short time, helping drivers or safety systems to avoid or mitigate earlier risk.

Supplemental Notes:

This paper was sponsored by TRB committee ACS20 Standing Committee on Safety Performance and Analysis.

Report/Paper Numbers:

TRBAM-23-02688

Language:

English

Corporate Authors:

Transportation Research Board

Authors:

Lyu, Nengchao
Wu, Jingcheng

Pagination:

6p

Publication Date:

2023

Conference:

Transportation Research Board 102nd Annual Meeting

Location: Washington DC, United States
Date: 2023-1-8 to 2023-1-12
Sponsors: Transportation Research Board

Media Type:

Web

Features:

Figures; References; Tables

Subject Areas:

Highways; Safety and Human Factors; Vehicles and Equipment

Source Data:

Transportation Research Board Annual Meeting 2023 Paper #TRBAM-23-02688

Files:

TRIS, TRB, ATRI

Created Date:

Feb 1 2023 10:33AM