Research On Recurrent Neural Networks For Short-Term Traffic Flow Analysis And Prediction

Sine the Opening policy has been taken,with China's booming growing economic,the per capita car ownership is increasing,and the construction of transportation network is also gradu ally developing.The increasing travel pressu re an d the increasing demand of ITS,short-term traffic flow forecasting is an important research direction of ITS.Accurate and the demand for daily routine,traffic development planning has been increaseing great significance.The main tasks of this thesis are as follows:(1)To deal with the massive traffic flow data used in the experiment in this thesis,completed the data cleaning and filling,and completed the necessary flow statistics and time series division for the modeling process,basing an important foundation for thesis research.(2)Implement traffic network node congestion prediction measure based on PageRank algorithm.This thesis finding the similarity of traffic network link structure and page structure,using this characteristic to introduce the formula to measure the traffic of the network node PageRank value.During the experimental process find t-1 time PageRank value has strong linear correlation with the traffic congestion index at mean time,using the correlation to predict traffic congestion has 83.6%of the average accuracy.(3)Building Recurrent Neural Net,works investigate the prediction of short-term traffic flow,and the traffic flow prediction experiment was based on the Vanilla LSTM model which using the single sampling point time flow sequence.At the same time,the performance of CNN-LSTM on the flow prediction using traffic flow matrix.This thesis also use the LSTM evolving GRU model on the traffic flow prediction were compared respectively.The experiment comprehensively compared the prediction errors of the three recurrent neural networks and the prediction error of the CNN-LSTM model was 30%lower than LSTM model and made great progress than the traditional model.(4)Build the short-term traffic flow analysis system based on Spark framework,the systems implement distributed computing and data storage through command line and functional programming.The PageRank traffic congestion prediction algorithm in chapter-3 is implemented on this system,and the system also has Spark SQL and Spark MLlib extension.