基于冰雪路面危险驾驶行为谱的行车风险识别方法

Driving Risk Identification Method Based on Dangerous Driving Behavior Spectrum on Ice and Snow Pavement

为描述驾驶人在冰雪路面行车时危险驾驶行为的具体表现形式,并量化其整体行车风险,通过模拟驾驶试验获取驾驶人操作类数据和车辆运动类数据,构建包含猛打方向、急变速、超速行驶、横向摇摆和跨道行驶这5种行为的危险驾驶行为谱。根据具体试验条件和四分位差法,分别确定不同路面附着系数下危险驾驶行为的阈值,并通过加权平均的方式计算危险驾驶行为谱特征值。以特征值作为行车风险评价指标,利用K-means聚类算法,将行车风险状态划分为4类,将随机森林模型、元启发式优化算法与支持向量机(Support Vector Machine, SVM)结合,建立冰雪路面行车风险识别模型。结果表明:超速行驶对冰雪路面行车安全影响最大,其权重为0.285;危险驾驶行为谱特征值主要分布于区间[0.00, 0.20]内,最大值出现在路面附着系数为0.2条件下;在路面附着系数为0.2和0.4条件下,严重行车风险明显增加,其占比分别为正常路面的4.85倍和2.49倍;鱼鹰优化算法(Osprey Optimization Algorithm, OOA)与SVM结合得到的行车风险识别模型的识别准确率为93.96%,优于浣熊优化算法(Coati Optimization Algorithm, COA)和灰狼优化算法(Gray Wolf Optimization Algorithm, GWOA)。研究结果有助于了解驾驶人在冰雪路面和正常路面行车时的差异,为提升驾驶人的行车安全水平提供依据。

To describe the specific forms of dangerous driving behavior and quantify the overall driving risk of drivers on ice and snow pavement,driving operation data and vehicle motion data were collected through simulated driving experiments.A dangerous driving behavior spectrum was constructed,which included five behaviors:sharp changes in direction,sharp changes in speed,speeding,lateral swaying,and occupying adjoining lanes.The threshold values of dangerous driving behavior under different pavement adhesion coefficients were determined based on specific experimental conditions and the quartile difference method.Then,the characteristic value of the dangerous driving behavior spectrum was calculated by weighted averaging.The characteristic value of the dangerous driving behavior spectrum was used as the driving risk evaluation index.The K-means clustering algorithm was used to divide the driving risk state into four categories,and the Random Forest model,metaheuristic optimization algorithm,and Support Vector Machine(SVM)were combined to establish a model for identifying driving risks on ice and snow pavement.The results show that speeding has the greatest impact on driving safety on ice and snow pavement with a weight of 0.285;the characteristic values of the dangerous driving behavior spectrum are mainly distributed in the interval[0.00,0.20],and the maximum value appears under the pavement adhesion coefficient of 0.2;under the conditions of pavement adhesion coefficients of 0.2 and 0.4,the severe driving risk increases significantly,accounting for 4.85 times and 2.49 times of normal pavement,respectively.The recognition accuracy of the driving risk recognition model obtained by combining the Osprey Optimization Algorithm(OOA)with SVM is 93.96%,outperforming the Coati Optimization Algorithm(COA)and the Grey Wolf Optimizer Algorithm(GWOA).The research results provide insight into the understand the differences between driving on ice and snow pavement and normal pavement,and provide a basis for enhancing driving safety.