Short Term Traffic Flow Forecasting Based On Integration Of LSTM Model And Grey Model

Road traffic system is the basis of a country’s national economic development,it is important to build a reasonable and efficient road traffic system.The object of this paper is traffic flow data,the purpose of this paper is to predict the traffic flow of a selected section of the next day.The main work of this paper is to preprocessing the traffic flow data,build the traffic flow predict model and test the prediction results.The main research contents and results are as follows:(1)The parameters,characteristics and influencing actors of traffic flow are analyzed,select time and traffic as parameters.EViews is used to obtain the seasonal and trend of data,then preprocessing the traffic flow data.First,we regulated the order of the data,added the missing data and corrected error data;Then performed wavelet threshold denoising,including the wavelet decomposition,the soft threshold denoising and the wavelet reconstruction,using the matlab code and get the data table and data picture after denoising.(2)Establish LSTM model and GM model of traffic flow data.LSTM model is built on the keras framework and uses python code.The traffic flow data is fed to the LSTM network,and determine the parameters and weights by studying the characteristics of the data.The prediction results of LSTM model are better,but the amount of data needed for model training is larger.GM belongs to grey model,we use a set of 10 data to dynamically change the model parameters and construct a dynamic grey model.The prediction result of GM model is not as good as that of LSTM model,but the amount of data needed is less and the real-time is strong.(3)Integrated LSTM model and GM using dynamic weights W.The single model prediction method is easy to be missed and ignored when dealing with the unexpected situation,which leads to the decrease of prediction accuracy.The two forecasting models are integrated to predict the traffic flow.We select the weight of the data with the highest correlation coefficient,the weights are consistent with the GM modeling steps.The prediction results show that the prediction accuracy of the integrated model is higher than that of the two models.