【摘要】 为研究L3自动驾驶事故场景下人工接管后换道轨迹的评价和分类问题，通过驾驶模拟实验采集换道轨迹数据；从舒适性、高效性、生态性、安全性4个方面选取9个评价指标；采用熵权TOPSIS（technique for order preference by similarity to an ideal solution）模型对换道轨迹进行评价并完成标签标定；用标定后的数据训练得到支持向量机（support vector machine, SVM）分类器模型，并将其应用于换道轨迹的分类中，该模型在测试集的平均准确率为79.55%,平均精确率为79.52%,平均召回率为79.51%,平均F₁值为77.43%。结果表明：应用熵权TOPSIS模型得到的评分最高的换道轨迹在舒适性、高效性、生态性和安全性上综合表现优秀；SVM分类器能以较为稳定的准确率完成换道轨迹的分类。得到的最优换道轨迹可为驾驶员的换道提供指导，也可为自动驾驶车辆的轨迹遵循提供参考。更多还原

【Abstract】 In order to study the evaluation and classification of lane-changing trajectory in L3 autonomous driving accident scenarios, the lane-changing trajectory data was collected through driving simulation experiments. Nine evaluation indicators were selected from the four aspects of comfort, efficiency, ecology, and safety. The entropy weight technique for order preference by similarity to an ideal solution（TOPSIS） model was used to evaluate the lane change trajectory and complete the label calibration. The calibrated data was trained to obtain the support vector machine（SVM） classifier model and applied it to the classification of lane-changing trajectories. The average accuracy of the model in the test set is 79.55%, the average accuracy is 79.52%, and the average recall is 79.51%, the average F₁ value is 77.43%. The results show that the highest-scoring lane-changing trajectory obtained by applying the entropy weight TOPSIS model has a comprehensive performance in comfort, efficiency, ecology and safety. The SVM classifier can complete the classification of the lane-changing trajectory with a relatively stable accuracy. The obtained optimal lane-changing trajectory can provide guidance for the driver to change lanes, and can also provide a reference for the trajectory following of the autonomous vehicle.更多还原