An Expressway Traffic Flow Prediction Model Based On GRU&Bi-LSTM

The 14 th Five-Year Plan of to make China with strong transportation network should continue to promote the expressway infrastructure construction,but improve the intelligent level of expressway construction and operation has become an important part.While providing high-speed,fast and economical traffic services,expressways are susceptible to traffic accident,weather,geographical conditions and holidays or time periods.Therefore,how to provide efficient,intelligent,unobstructed and safe intelligent expressway environment,and provide timely and accurate traffic information services for travelers has become an urgent problem for researchers.Taking the expressway traffic flow prediction as the research goal,the expressway toll database as the research basis,combineing with the weather database,this paper puts forward an expressway traffic flow prediction model based on GRU&Bi-LSTM.This paper proposes a GRA-S&K data analysis method to solve the problems that the factors affecting the expressway traffic flow are complex and varied,and the input-data of the expressway traffic flow prediction model is not fully processed.This method accurately analyzes the multiple influencing factors of the expressway traffic flow,eliminates the weak influencing factors,reduces the dimension of the input-data of the prediction model,and obtains the dynamic weight allocation of the influencing factors,which improves the training efficiency of the model and the accuracy of the prediction.Through experimental comparison,the results show that the prediction root mean square error of the data processed by GRA-S&K is 40.51% lower than that of the data not prosessed,and mean absolute error of the data processed by GRA-S&K is 37.36% lower than that of the data not prosessed.In view of the time regularity of expressway traffic flow,this paper puts forward an expressway traffic flow prediction model based on GRU&Bi-LSTM.Adam is used to update weights in the network,and the model parameters are optimized by integrating convergence and prediction results.The model retains the advantages of LSTM,which can save long dependent relationships,and combines the characteristics of time-series data,at the same time,taking into account the impact of the changes in traffic flow before and after,and takes advantage of GRU that the structure is simple enough to ensure the high learning efficiency of the model.Finally,taking the expressway toll of Shaanxi Province as the experimental target,the traffic flow of three representative toll stations in Guanzhong Plain,Loess Plateau and Qinba mountain area is studied.The results show that compared with the GRU,Bi-LSTM and Bi-LSTM&Gru composite model,the prediction accuracy of the expressway traffic flow prediction model based on GRU&Bi-LSTM is significantly improved,and it has certain universality for multi-type terrain and multi-influencing factors.