基于SVM的危货车驾驶员饮水分心判别模型

Discriminant Model for Dangerous Goods Vehicle Driver Drinking Water Distraction Based on SVM

执行车内次任务是导致驾驶员注意力分散乃至造成交通事故的重要原因。为对危货车驾驶员饮水分心进行辨识,进行了驾驶模拟试验。试验中,将水杯放置在危货车驾驶舱内3种常见的杯架位置处,采集饮水分心驾驶和正常驾驶时的车辆运行状态数据。在差异性统计分析提取特征指标的基础上,对变化较快的特征引入频域分析方法提取特征指标,经分析新增加的特征指标能够反映被试驾驶员对车辆横向稳定性的控制能力。使用支持向量机建立驾驶员饮水分心驾驶判别模型,分别使用遗传算法和粒子群算法对初步判别模型进行优化并对比分析,最终对优化后的模型进行评估。结果表明,应用遗传算法和粒子群算法优化后的判别模型对驾驶员饮水分心状态的正确分类比例分别达到94.02%和93.21%,遗传算法的参数寻优结果优于粒子群算法,同时遗传算法主要通过选择、交叉、变异操作产生新的种群个体进化到下一代,是较为成熟的收敛性分析方法,故认为遗传算法更适合作为驾驶员饮水分心驾驶判别模型的优化算法。

The implementation of the secondary task is the important reason leading to distract the driver and even cause traffic accidents. In order to identify the distracted driving of dangerous goods vehicle driver,a driving simulation test is carried out. In the test,the water cup is placed in 3 common cup holders in driver's cabin,and the data of the driving condition of the distracted driving and normal driving are collected. On the basis of difference statistical analysis to extract the characteristic indexes,introducing the frequency domain analysis method to extract the characteristic indexes which change faster,the analysis shows that the new added indexes can reflect the tested driver's ability to control lateral stability of vehicle. The discriminant model of driver drinking water driving is established using support vector machine( SVM),the preliminary discriminant model is optimized using GA and PSO,which are comparative analysed. Finally,the optimized model is evaluated. The result shows that using the optimized discriminant models using GA and PSO,the correct classification rates of driver drinking water distraction state reaches 94. 42% and 94. 08%respectively. GS is superior to PSO in parameter optimization. generates new population individuals to evolve to the next generation by selection,crossover and mutation operation,it is a more mature method of convergence analysis. So it is believed that GA is more suitable as the optimization algorithm to the driver driving water distraction discriminant model.