Research On Short-term Traffic Flow Prediction Based On Deep Learning

With the development of the city and the improvement of people’s living standard,the number of private vehicles is increasing,the traffic volume of urban traffic is increasing rapidly,and the traffic load is increasing day by day.The intelligent transportation system(its)can optimize the traffic efficiency under the condition of original traffic hardware by optimizing the process of traffic dispatch and control,and realize the effect of " Ease congestion and keep traffic flowing".But for the intelligent transportation system,whether the signal timing optimization or the route guidance planning,all depend on the traffic flow data.it can’t get good results by using real-time traffic flow data to control the next period of traffic,so the short-term traffic flow prediction has become a research hotspot in the field of traffic.In addition to being highly stochastic and nonlinear,the traffic flow data also have the characteristics of periodicity,space-time similarity and so on.Deep Learning,as a method that can represent high level abstract attributes or features by combining low level features,can make deep neural networks that stack multiple hidden layers have better features learning ability and distributed expression ability,it does a very good job of characterizing the implicit laws within the data.Therefore,the deep learning method and model have natural advantages to solve the short-term traffic flow forecasting problem.The main research work of this paper is as follows:1 First of all,this thesis explains the background and significance of the research project,summarizes the current common short-term traffic flow forecasting models,and elaborates the working principle and development process of deep learning technology,the popular deep learning models and their applications in short-term traffic flow forecasting are analyzed and summarized2 The concept of short-term traffic flow forecasting is introduced in detail,and the data collection and processing of traffic flow are described,and the whole process of traffic forecasting problem description,general forecasting model construction,training and evaluation.3 For short-term traffic flow forecasting problem,the Mianyang data type is analyzed,this paper introduces the data preprocessing before the model training,including terrain matching,path linking,data integration,graph structure building and Adjacency Matrix building,etc.,in this paper,a method of missing path interpolation and a graph data reduction method for short-term traffic flow prediction are presented.4 To solve the problem of short-term traffic flow prediction,the substructures of Graph CNN-LSTM model,including Graph CNN,LSTM and FFNN,are introduced.This paper mainly introduces two kinds of deformation of Graph CNN and the influence of the deformation on the whole model.At last,the experiment is set up and the experiment is carried out.Then the experiment results are analyzed and the influence of different models on short-term traffic flow prediction is analyzed.The different functions and advantages of the methods used in the study in short-term traffic flow forecasting are obtained.