基于深度学习组合模型的交通流预测  

Traffic flow prediction based on deep learning combined model

作　　者：郅英冲 赵金宝[2,3] 李晓飞 韩刚 孔维超 潘春雨 ZHI Ying-chong;ZHAO Jin-bao;LI Xiao-fei;HAN Gang;KONG Wei-chao;PAN Chun-yu(Jinan Rail Transit Group Co.,Ltd.,Jinan 250000,China;School of Transportation and Vehicle Engineering,Shandong University of Technology,Zibo 255000,China;School of Transportation,Southeast University,Nanjing 210009,China)

机构地区：[1]济南轨道交通集团有限公司,山东济南250000 [2]山东理工大学交通与车辆工程学院,山东淄博255000 [3]东南大学交通学院,江苏南京210009

出　　处：《广西大学学报（自然科学版）》2022年第3期832-841,共10页Journal of Guangxi University（Natural Science Edition）

基　　金：国家自然科学基金项目(51608313);山东省自然科学基金项目(ZR2021MF109);山东高速集团科技项目(2020-SDHS-GSJT-024);山东省交通运输厅科技计划项目(2019B11);济南市科技计划项目(201807005)。

摘　　要：为了研究高速公路在交通流预测过程中时间粒度对于精度的影响,及时向出行者提供精准实时的道路信息,通过TensorFlow深度学习模块,建立支持向量回归(SVR)与长短时记忆(LSTM)相结合的预测模型。基于LSTM模型的长期记忆功能与支持向量回归非线性化特点,调整优化参数,以英国高速公路局提供的M3高速公路数据为例进行分析,根据预测结果对模型进行对比评价,并结合模型对不同时段的交通流数据进行分析研究。结果表明:SVR-LSTM组合预测模型对于高速公路数据有更好的适应性,而且时间粒度如果越精细,预测精度将大幅提高。In order to study the impact of time granularity on the accuracy of expressway in the process of traffic flow prediction,and to provide travelers with accurate(SVR)and real-time road information in time,a prediction model combining support vector regression and long short-term memory were established through the TensorFlow deep learning module.Based on the model of LSTM with long-term memory function and the non-linear characteristics of support vector regression,the parameters were continuously optimized.The M3 highway data provided by the British Highway Administration was used as an example for analysis and verification.The model was compared and evaluated based on the prediction results,and we also analyzed the traffic flow data at different time periods combined with the model.The experimental results show that SVR-LSTM combined prediction model has better adaptability to highway data,and the finer the time division is the greater the prediction accuracy will be improved.

关 键 词：交通数据 交通流预测 深度学习 支持向量回归 长短期记忆 

分 类 号：U491[交通运输工程—交通运输规划与管理]