基于FCNN神经网络的多驾驶风格车辆换道预测

Prediction of lane changing with multiple driving styles for intelligent vehicles based on FCNN neural network

在未来复杂的智能交通环境中,自动驾驶车辆与人类驾驶车辆混合共享道路成为可能。在这种混合交通流的背景下,准确预测人类车辆的换道行为对于自动驾驶系统的安全性和效率至关重要。本研究针对车辆换道的主要影响因素—周围环境和驾驶风格,提出了一种基于FCNN神经网络的人工车辆多驾驶风格换道决策预测方法。首先,通过应用K均值聚类算法对驾驶员的驾驶风格进行分类,从而综合考虑驾驶风格的差异。然后,将分类结果作为FCNN神经网络的输入变量,用于预测人类车辆的换道行为。该方法基于真实车辆轨迹构建的下一代仿真数据集进行训练和测试,并取得了91.80%的准确率。实验结果表明,通过结合驾驶风格分类和FCNN神经网络,能够更准确地预测人类车辆在不同驾驶风格下的换道行为。本研究的成果对于混合交通流下自动驾驶车辆预测人类车辆的换道行为具有重要的实际应用价值,并为未来进一步改进和拓展提供了新的思路。

In the future complex intelligent traffic environment,the coexistence of autonomous vehicles and human-driven vehicles on the road becomes a possibility.In the context of this mixed traffic flow,accurately predicting the lane-changing behavior of human-driven vehicles is crucial for the safety and efficiency of autonomous driving systems.This study proposes a prediction method for lane-changing decisions of human-driven vehicles with multiple driving styles,based on the FCNN(Fully Convolutional Neural Network),focusing on the key factors of lane-changing:the surrounding environment and driving styles.Firstly,by applying the K-means clustering algorithm to classify driving styles of drivers,the variations in driving styles are comprehensively considered.Subsequently,the classified results are utilized as input variables for the FCNN neural network to predict the lane-changing behavior of human-driven vehicles.The method is trained and tested on a next-generation simulation dataset constructed from real vehicle trajectories,achieving an accuracy rate of 91.80%.Experimental results demonstrate that by combining driving style classification and FCNN neural network,it is possible to more accurately predict the lane-changing behavior of human-driven vehicles under different driving styles.The findings of this research is of significant practical value for predicting the lane-changing behavior of human-driven vehicles in the context of mixed traffic flow with autonomous vehicles.Additionally,it provides new insights for future improvement and expansion of this approach.