Short Term Traffic Flow Prediction Of Expressway Sections And Urban Intersections Based On Deep Learning

With the continuous development of society and the progress of the times,people’s living standards have been greatly improved.The number of private vehicles in China is increasing The urban traffic flow is growing rapidly.For intelligent transportation system,the signal timing optimization and the travel path guidance planning all depends on the traffic flow data during the optimization operation.Using the real-time traffic data to control the traffic in the next period often can not get good results,so the short-term prediction of traffic flow has become a research hotspot in the field of transportation.Deep learning is a method of combining multiple hidden layers of deep neural network with low-level features to express high-level abstract attributes or features.It can well describe the hidden rules in data.Therefore,deep learning method and model can well mine various underlying features in traffic flow data,and use these features to predict future traffic situation.It has great advantages to use deep learning to predict short-term traffic flow.This paper uses Python crawler technology to crawl PEMS-I105 data set in California,USA.In this paper,convolution neural network,long short-tern neural network,bidirectional long short-term neural network,gated recurrent units,bidirectional gated recurrent units,bidirectional long gated graph convolutional neural network(BLG-GCN),spatial time graph convolutional neural network for traffic(S2TGCN),spatial time graph convolutional neural network based on attention for traffic(AS2TGCN)were used to predict the short-term traffic volume in a main road in a Los Angeles City’s expressway in PEMS-I105 datasets.The above methods were also used in the prediction of traffic volume at the intersection of Huangshan Road and Tianzhi road in Hefei City,Anhui Province.The paper will change the number of neurons in the hybrid recurrent neural network model,the number of convolution cores in the convolution layer of CNN2D model,the convolution core size in the convolution layer of CNN2D model,the number of neurons in the graph convolution layer of AS2TGCN model,the dropout layer ratio of BLG-GCN model,the learning rate of BLG-GCN model,the deviation of S2TGCN model,the optimizer of BLG-GCN model,and the recurrent_dropout parameter of hybrid recurrent neural network model,the weight initialization method of hybrid recurrent neural network and regularization of hybrid recurrent neural network applied on weight.The comparison of performance changes are made.And the paper will select the optimal performance parameter.The performance of the above methods and models are compared and analyzed to select the optimal solution based on various performance indicators and identify the best short-term traffic flow prediction method suitable for one-direction Expressway and each entrance of intersection.