A Recurrent Neural Network Approach Using LSTM

Abstract

In order to traffic safety, energy control and environmental protection forecasting the traffic flow is considerably significant. Still predicting accurate traffic flow is really challenging task. In recent years the data has been increased vastly with the evolution of smart cities, this indicates that we have set foot in the era of enormous information for transportation. And this advancement is directly affecting transportation networks, thus decreasing the travel time leads to increase efficiency and reduces the bad impact of vehicles on the environment. Now the foremost issue of traffic flow forecasting is the sharp nonlinearities because of change between congestion, free flow, breakdown and recovery. We present that the deep learning architecture can process these nonlinear effects. This study has used previous traffic data to predict short term traffic flow; the collected data is real-time traffic data. The proposed model includes many traffic characteristics like: speed, density, volume, time and day of weeks as input variables. In this paper, we have proposed a new model, called ILSTM, to process nonlinear data and to attain more precise traffic flow prediction. It takes advantages of LSTM model.