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

With the rapid development of society and economy,the number of motor vehicles has increased rapidly,and urban traffic congestion has become increasingly serious.In order to solve the problem of traffic congestion,transportation departments have widely applied intelligent transportation systems for traffic management.Short-term traffic flow prediction is not only one of the core contents of the intelligent transportation system,but also the basic basis for the implementation of traffic control and traffic guidance by transportation departments.How to accurately predict short-term traffic flow has been a hot issue for scholars in various countries.Short-term traffic flow has the characteristics of randomness,periodicity,correlation,and non-linearity.It is a typical time series data.Accurately extracting the time-series characteristics of traffic flow is the key to improving the short-term traffic flow prediction model.In recent years,deep learning methods have gradually been applied in traffic flow prediction because of their strong nonlinear fitting ability and deep feature expression ability of data.This paper studies the application of Long Short-Term Memory(LSTM)algorithm and Gated Recurrent Unit(GRU)algorithm with time memory characteristics in deep learning to predict short-term traffic flow.The main contents are as follows:(1)Aiming at the time-series characteristics of short-term traffic flow data,this paper uses LSTM algorithm and GRU algorithm with temporal memory characteristics in deep learning to study short-term traffic flow prediction.And the short-term traffic flow prediction model based on LSTM and GRU were established respectively.The network structure and parameters of the prediction model were selected through experimental comparisons.Finally,experiment results show that these two short-term traffic flow prediction models can better find the time-series characteristics of traffic flow data and can predict the future traffic flow more accurately.(2)In order to further improve the accuracy of the short-term traffic flow prediction model based on LSTM and GRU,the attention mechanism method in deep learning is used,and the LSTM-attention and GRU-attention short-term traffic flow prediction models are established respectively.By repeatedly using the Attention mechanism to calculate multiple hidden units in the model and actively paying attention to the key information in traffic flow prediction,the entire network structure can adaptively assign weights,thereby improving the accuracy of the traffic flow prediction model.The comparison experiment shows that the prediction accuracy of the improved prediction model using Attention is greatly improved compared with the original model.(3)Aiming at the problems of low prediction accuracy of single traffic flow prediction model and easy to produce overfitting,the optimal weighting method was used to establish two types of short-term traffic flow combination prediction models.One is a combination prediction model based on LSTM and GRU,and the other is a combination prediction model based on LSTM-attention and GRU-attention.The comparison experiment results between the combined prediction model and the single prediction model show that the prediction accuracy of the two combined prediction models proposed in this paper is higher than their original single prediction model.