基于决策机理与支持向量机的车辆换道决策模型

驾驶决策机制是保障自动驾驶车辆驾驶安全的关键技术,而换道研究是其重要课题. 然而,在复杂的动态环境下行驶时,使智能车辆做出安全、符合要求的换道决策仍然是一个难点. 为此,首先分析了车辆自由换道的影响因素,采用传统的数理模型建立了基于换道收益、安全和必要性的车辆换道规则模型. 其次,针对在不同的驾驶工况换道决策考虑的因素不同,提出从基于物理状态的特征、基于交互感知的特征以及基于道路结构的特征三个方面提取决策变量,使换道模型决策时考虑的因素更加全面. 然后,针对自由换道决策过程中存在的多参数和非线性问题,提出了基于贝叶斯优化算法（BOA）的支持向量机（SVM）决策模型. 最后,所提出的模型在NGSIM数据集上进行验证,对比试验表明:建立的BOA Gaussian-SVM模型具有较高的综合预测性能,对换道行为的识别准确率可达到92.97%,超越了其他模型并远高于规则模型. 同时在Airsim平台上进行了仿真实验,实验结果进一步证明了BOA Gaussian-SVM决策模型的有效性,说明此模型可进一步应用到自动驾驶和辅助驾驶系统开发中.

Vehicle lane-changing decision model based on decision mechanism and support vector machine

This paper first analyzes the influencing factors of free lane change of autonomous driving vehicle, and uses the traditional mathematical model to establish a vehicle lane change rule model based on the benefits, safety and necessity of lane change. Second, in view of the different factors considered in lane changing decision-making under different driving conditions, this paper proposes to extract decision variables from three aspects: physics-based features, interaction-aware features and road-structure-based features, and designs a feature extraction algorithm to make the factors considered in lane changing model decision-making more comprehensive. Then, for the multi-parameter and non-linearity problems existing in the decision-making process of autonomous lane change, a support vector machine (SVM) decision-making model based on Bayesian optimization algorithm (BOA) is proposed. Finally, the proposed model is verified on the NGSIM data set. The comparison test shows that the established BOA Gaussian-SVM model has a high comprehensive prediction performance, and the recognition rate of channel change behavior can reach 92.97%, which is better than other models and much higher than rule-based model. At the same time, simulation experiments are carried out on Airsim platform, and the results prove the effectiveness of BOA Gaussian-SVM decision model.