考虑随机拥堵情况的快速路交通流短时预测方法研究  

Research on Short-time Prediction Method of Expressway Traffic Flow considering Stochastic Congestion Scenario

作　　者：金梦媛 JIN Mengyuan(Zhejiang Institute of Communication Co.,Ltd.,Hangzhou Zhejiang 310031,China)

机构地区：[1]浙江数智交院科技股份有限公司,浙江杭州310031

出　　处：《交通节能与环保》2024年第3期40-44,61,共6页Transport Energy Conservation & Environmental Protection

摘　　要：本研究采用苏州快速路多路段交通数据,基于现有多种短时交通流预测方法,采用二次指数平滑模型、LSTM神经网络模型和BP神经网络模型三种不同模型建立车速预测模型进行对比研究,用MSE辅之以实验数据可视化作为模型评判基准,探究不同模型在随机拥堵情况下的预测性能。结果显示,在交通流平稳运行情况下,三者预测性能良好,MSE值均在30左右。但在随机短时拥堵情况下,LSTM和二次指数平滑模型的预测性能下降,而BP模型在处理车速时段性的不确定变化时展现出更好的性能,训练结果 MSE值在20左右,应对突发情况时MSE值也能维持在40以下,相比指数平滑模型和LSTM模型具有更好的预测性能。In this study,traffic data of multiple sections of Suzhou expressway were used.Based on various existing short-term traffic flow prediction methods,the vehicle speed prediction model was established by using three different models,namely,secondary exponential smoothing model,LSTM neural network model and BP neural network model,and MSE supplemented with experimental data visualization was used as the model evaluation benchmark to explore the prediction performance of different models under random congestion.The results show that when the traffic flow runs smoothly,the three prediction performance is good,and the MSE value is about 30.However,in the case of random short-term congestion,the predictive performance of LSTM and quadratic exponential smoothing model declines,while the BP model shows better performance when dealing with uncertain changes in a certain period of time.The MSE value of the training result is around 20,and the MSE value can also be maintained below 40 when dealing with emergencies.Compared with exponential smoothing model and LSTM model,BP neural network model has better predictive performance.

关 键 词：交通碳排放 交通流预测 快速路 指数平滑 LSTM BP神经网络 

分 类 号：U491.112[交通运输工程—交通运输规划与管理]