基于LSTM的跨海桥梁危化品车辆行驶轨迹预测

Trajectory Prediction of Hazardous Chemical Vehicles for Sea-crossing Bridges Based on LSTM

我国沿海地区气象环境复杂,跨海桥梁上车辆混杂密集,车流量大、车辆混杂和侧风环境等因素都对跨海桥梁危化品车辆的安全行驶产生了极大的影响,因此跨海桥梁危化品车辆行驶轨迹研究对于加强危化品运输安全管理具有重要的现实意义。针对跨海桥梁危化品运输车辆的轨迹预测问题,利用深度学习方法,建立了一种基于长短时记忆网络(LSTM)与注意力机制的跨海桥梁危化品车辆行驶轨迹预测模型,该预测模型包括信息输入模块、注意力层和轨迹输出模块,并利用沿海一座跨海大桥上危化品车辆行驶环境实测轨迹数据对预测模型的预测性能进行了分析与验证。结果表明:LSTM能适应长时域的车辆行驶轨迹预测,注意力机制提高了预测模型的训练速度;由于考虑了车辆间交互影响因素,该预测模型具有更高的准确性和计算效率,同时LSTM也减少了预测模型在时域较长情况下车辆行驶轨迹预测的误差。

The meteorological environment in China coastal area is complicated,and the vehicles on the seacrossing bridges are mixed and dense.The factors such as large traffic flow,mixed vehicles and crosswind environment have a great influence on the safe driving of hazardous chemical vehicles on sea-crossing bridg-es.Therefore,the study of hazardous chemical vehicle trajectory for sea-crossing bridges has important practical significance for strengthening the safety management of hazardous chemical transportatioru In order to achieve the goal of trajectory prediction of hazardous chemical vehicles on sea-crossing bridges,a trajectory prediction model of hazardous chemical transport vehicles on sea-crossing bridges is established based on long short-term memory network(LSTM)and attentional mechanism by using deep learning algorithm.The model is mainly composed of information input module,attention layer and trajectory prediction output module.At the same time,the model is trained and verified by using the measured track data of hazardous chemical vehicle driving environment on a cross-sea bridge・The results show that the LSTM can adapt to the trajectory prediction in long time domain,and the attentional mechanism improves the training speed of the model.Due to the consideration of the interaction between vehicles and the influence of cross-wind environment,the prediction model with attentional mechanism has higher accuracy and computational efficiency,and LSTM also reduces the vehicle trajectory prediction error in the longer time domain.