基于理论驱动和数据驱动的组合跟车模型  

Research on data-driven and theory-driven hybrid car-following model

作　　者：葛世磊 霍为炜 龚国庆[1] GE Shilei;HUO Weiwei;GONG Guoqing(School of Mechanical and Electrical Engineering,Beijing Information Science&Technology University,Beijing 100192,China)

机构地区：[1]北京信息科技大学机电工程学院,北京100192

出　　处：《重庆理工大学学报（自然科学）》2024年第11期43-51,共9页Journal of Chongqing University of Technology：Natural Science

基　　金：国家自然科学基金面上项目(52077007,52172354)。

摘　　要：基于理论驱动的模型虽成功解释了观察到的交通行为,但无法处理多变的驾驶行为信息,导致模型预测能力较差。基于数据驱动的深度学习预测模型能够处理复杂的驾驶信息,但需要大量的驾驶数据进行模型训练。针对2类模型存在的问题,提出一种结合模型,将理论驱动模型(OV)与数据驱动模型(PSO-CNN-BiLSTM-Att)相结合,形成组合跟车模型,将IDM模型和PSO-CNN-BiLSTM-Att神经网络的预测结果相结合,这种融合保留了理论驱动模型提供的可控性,同时也利用了数据驱动模型的预测精度。通过NGSIM交通数据,与单独的OV理论驱动模型和PSO-CNN-BiLSTM-Att数据驱动模型相比,组合模型的预测误差显著减少,分别降低了88%和67%。此外,还进行了不同驾驶行为风格模拟,结果表明组合模型可以真实反映跟车行为。Autonomous driving has been advancing rapidly and is poised to replace human drivers in the future.One of the crucial tasks in realizing fully autonomous driving is accurately predicting the following vehicles’behaviors,whose models have been extensively researched.Existing models are categorized into theory-driven models and data-driven models.Theory-driven models,such as the Optimal velocity Model(OV),successfully explain observed traffic behaviors but struggle to handle the varied information of driving behaviors,resulting in poor speed prediction capabilities.Data-driven deep learning prediction models can handle complex driving information but require extensive driving data for model training.To address the limitations of both types of models,we propose a hybrid model integrating the theory-driven model(IDM)with the data-driven model(PSO-CNN-BiLSTM-Att).By integrating the prediction results of the OV model and the PSO-CNN-BiLSTM-Att neural network,our model preserves the controllability provided by the theory-driven model while leveraging the prediction accuracy of the data-driven model.Utilizing NGSIM traffic data,it markedly reduces the prediction error by 88%and 67%respectively compared to those of individual OV theory-driven and PSO-CNN-BiLSTM-Att data-driven models.Meanwhile,our simulations of asymmetric driving behavior styles demonstrate our hybrid model accurately monitors following vehicles’behaviors.

关 键 词：跟车行为 最优速度模型 长短期记忆网络 粒子群优化 组合预测 

分 类 号：U463.6[机械工程—车辆工程]