Research On Short-term Traffic Flow Forecasting Based On LSTM And GRU Entropy Weight Integration

With the development of economy and the popularization of vehicles,while facilitating people’s travel,traffic problems such as road congestion and road pressure are also brought about.For this reason,we put forward an intelligent transportation system,which can effectively relieve road pressure and facilitate people to travel by scientific methods to manage the traffic system,accurately predict traffic flow and timely make correct traffic guidance.Therefore,the accurate prediction of short-term traffic flow is the most important part of scientific traffic management.For many years,people have been trying to solve the problem of short-term traffic flow forecasting,put forward various schemes,and have abundant practice in traffic management.Under the dual demand of scientific and technological progress and economic development,accurate prediction of traffic flow poses new challenges to people.Aiming at the typical time series characteristics of traffic flow sequence,this paper designs two different kinds of recurrent neural networks(RNN):Long Short Term Memory(LSTM)network and Gated Recurrent Unit(GRU)network.LSTM and GRU algorithm can increase the memory function of time series and can be passed on very well.The feature continuity of traffic flow history data is presented,and the feature extraction of traffic flow history data is carried out.The traffic flow in the next 24 hours is predicted with 5 minutes as the time interval.Through experimental training,the optimal structure of cyclic neural network is obtained to solve the problem of short-term traffic flow forecasting.Compared with the single hidden layer structure,the LSTM and GRU errors of double hidden layer are reduced and the fitting degree is increased.When the fluctuation of data is large,LSTM and GRU algorithms with double hidden layers are not very accurate for peak prediction,and they are prone to over-fitting.In order to solve this problem,this paper uses the combination neural network method of entropy weight set to calculate the information entropy of each algorithm.According to the amount of information of the whole system,LSTM and GRU algorithms are integrated through the entropy weight distribution,and a new entropy weight distribution algorithm is obtained.Experiments show that the short-term traffic flow prediction algorithm based on the entropy weight method has higher accuracy and smaller error than the single LSTM or GRU algorithm,and the predicted value is closer to the peak value of the wave and better for the fluctuation prediction.