基于ETC门架数据的高速公路短时交通流预测

Predicting Short-term Traffic Flow on Expressway Based on ETC Gantry System Data

截止到2021年2月底,全国建设了2.66万套ETC门架系统,产生了大量数据,这些数据记录了断面、路段及路网的交通流参数,为交通流的预测提供了新思路,但是目前缺乏有效的应用。为合理利用ETC门架数据、探究基于ETC门架数据进行交通流预测的较优方法,选取了4种神经网络模型进行了预测结果和精度的对比。首先分析了ETC门架系统的构成及数据内容,研究了BP,ELMAN,RBF,GR神经网络模型的预测过程;在ETC门架系统产生的1T数据中,抽取6 d共231657条原始断面数据进行预处理,共得到3456条流量和速度数据。然后,将前4 d的数据作为4种模型的输入参数,在相同交通流时间序列、相同交通流时间汇集度和不同交通流状态向量维度下,预测未来2 d交通流的速度和流量。最后进行神经网络预测结果对比分析。结果表明:(1)ELMAN模型的预测精度最优,模型的各类评价指标(RMSE,MAPE,MAE)均有显著优势;(2)ELMAN神经网络模型预测速度比预测流量准确,在二者均能反映路网交通运行状态的情况下,可以由速度的预测结果反推流量的预测值。本研究将对高速公路交通流预测、评估等工作起到重要作用。

By the end of February 2021,26600 ETC gantry systems have been built nationwide,generating a large amount of data.These data record the traffic flow parameters of sections,road sections and road networks,providing new ideas for traffic flow prediction,but currently there is a lack of effective applications.In order to make reasonable use of ETC gantry system data and explore a better method of traffic flow prediction based on ETC gantry system data,4 neural network models are selected to compare the prediction results and accuracies.First,the system structure and data content of the ETC gantry system are analyzed,and the prediction process of neural network models of BP,ELMAN,RBF and GR are studied.In the 1 T data generated by ETC gantry systems,231657 pieces of original section data for 6 days are extracted for pre-processing,and 3456 pieces of volume and speed data are obtained totally.Then,under the same traffic flow time series,the same traffic flow time convergence degree and different traffic flow state vector dimensions,taking the data of the first 4 days as the input parameters of the 4 models,the traffic volume and speed in the next 2 days are predicted.Finally,a comparative analysis of neural network prediction results is conducted.The result shows that(1)The ELMAN model has the best prediction accuracy,and all kinds of evaluation indicators(RMSE,MAPE and MAE)have significant advantages;(2)The ELMAN neural network model is more accurate in predicting the speed than the volume,in the case that both of them can reflect the traffic state of road network,the predicted value of the volume can be back-deduced from the predicted result of the speed.This study will play an important role in expressway traffic flow prediction and evaluation.