Traffic Flow Prediction Of City Road Network Based On Deep Learning

Along with China's modernization process,China's social and economic development has been developing rapidly.By the end of 2017,the number of motor vehicles in China has reached 310 million.However,due to the aggregation of population into cities,the traffic network in cities is under tremendous pressure.Trafic congestion has become a common problem in urban road networks.In the current intelligent transportation information system,traffic flow prediction is an important part and the key to solve the problem of traffic congestion.On the basis of collecting a large amount of traffic information,accurate traffic flow prediction can help the intelligent transportation information system to analyze the road network efficiently and timely feedback traffic information to managers and travelers,so as to better plan the road network and improve the utilization ratio of road network.The traditional traffic flow prediction method is generally based on the nonlinear fitting of historical data of a single road.The data used in the traffic flow are small,the model is simple,and the overfitting is serious.Some scholars have taken into account the temporal and spatial characteristics of the road,and put forward the idea of road network compression.However,due to the large number of manual analysis characteristics,the generalization performance of the road network compression algorithm is poor,and it cannot support large-scale high dimensions.Urban road network data.In order to make up for the shortcomings of the above traditional traffic flow forecasting methods and better predict the large-scale urban road network as a whole,this paper proposes a deep learning based urban road network traffic flow prediction framework.Firstly,the feature engineering method in machine learning is introduced to deal with the characteristics of large scale road network data.The abnormal nodes in data are eliminated by anomaly detection method based on SVDD and isolated forest.The optimal parameters are obtained through the stationarity test based on difference,and the road network data is compressed by means of spectral clustering road network compression method based on CH index optimization.Secondly,based on LSTM and SAEs,the traffic flow prediction fusion model is designed and implemented.It can directly input data from feature engineering and road network compression processing,use the parameters provided by stationarity analysis,extract the shallow sequence characteristics of the data with stacked LSTM model,make the SAEs model self-encode the data,extract deep hidden features,and use dynamic weight.The fusion method integrates the two,taking into account the shallow and deep features of the data,which improves the prediction accuracy and generalization ability of the model.Finally,this paper uses the twin city traffic data of the traffic data research laboratory of University of Minnesota to conduct an experimental analysis of the framework.The experiments show that:(1)the fusion model based on SVDD and isolated forest can effectively identify and eliminate the bad points in the mass network with insufficient data and improve the quality of data;(2)spectral clustering based on CH index can compress the road network.In order to effectively compress the urban road network data,the compression rate reaches 20%under the premise of guaranteeing the prediction accuracy.(3)the fusion model prediction model based on depth learning proposed in this paper is superior to other models in prediction accuracy and performance,and the average precision of the whole network is 97.7%.