Abstract: Aiming at the problem that the rollover of heavy vehicles couldnt be predicted accurately under complicated driving conditions, an AdaBoost algorithm based on machine learning was designed, which realized the real-time and accurate calculation of non-trip rollover criterion of heavy vehicles under complex driving conditions. Firstly, the heavy vehicle simulation model and rollover warning model were established. Secondly, based on the AdaBoost learning algorithm theory, the architecture of multiple weak classifiers based on the single-layer decision method was designed, and the simulation training and weighted summation were carried out. Finally, combined with commercial software TruckSim dynamics software, the rollover effect of heavy vehicle rollover warning failure under double lane change (DLC) and fishhook conditions was compared and analyzed. The simulation results show that the proposed rollover warning criterion based on AdaBoost algorithm can effectively predict heavy vehicle rollover under complex driving conditions, the accuracy of the corresponding test set is 24.9% better than that of Logistic regression algorithm, and the receiver operation characteristic (ROC) curve area is 0.958.

Key words: vehicle engineering; rollover warning model; AdaBoost algorithm; single decision tree; load transfer ratio; heavy vehicles

摘要： 为有效解决复杂行驶工况下无法准确预测重型车辆侧翻的难题，设计了基于机器学习方法的自适应提升（AdaBoost）算法，实现了复杂行驶工况下重型车辆非绊倒型侧翻判据的实时准确计算。首先建立了基于重型车辆仿真模型与侧翻预警模型；其次，利用AdaBoost学习算法理论，设计了基于单层决策方法构建多个弱分类器的架构并对其进行了模拟训练与加权求和；最后，结合商业软件TruckSim动力学软件，对比分析了双移线（DLC）与鱼钩（Fishhook）工况下重型车辆侧翻预警失效的侧翻效果。仿真结果表明：所设计的基于AdaBoost算法侧翻预警判据可在复杂行驶工况下有效预测重型车辆侧翻，且对应的测试集正确率比Logistic回归算法预测精度改善24.9%，且模型评估预测ROC（receiver operation characteristic）曲线面积为0.958。

关键词: 车辆工程；侧翻预警模型；AdaBoost算法；单层决策树；载荷转移率；重型车辆