Short-term Traffic Flow Prediction Based On LSTM Deep Neural Networks

In order to solve the increasingly serious traffic problems,especially the problem of traffic congestion,traffic administration began to widely use intelligent transportation system(ITS)for dynamic traffic management.Short-term traffic flow prediction,is not only a core component of intelligent transportation system,but also the core basis for the implementation of traffic management and guidance by traffic administrative department.At the same time,accurate traffic flow prediction information can also provide detailed and real-time road information to improve the road capacity and avoid congestion.Therefore,accurate traffic flow prediction model is particularly important.In addition,with the continuous development and application of big data,more and more traffic flow data are obtained.How to use these data to predict short-term traffic flow more accurately has become a crucial issue.Due to the characteristics of uncertainty,periodicity,correlation and non-linearity,short-term traffic flow is a typical time-series data.The core of accurate prediction is to obtain the potential correlation and influence among data.Based on the above background,this thesis established a short-term traffic flow prediction model based on long-short term memory(LSTM).The specific research contents are as follows:Firstly,this thesis systematically analyzes the advantages and disadvantages of the current short-term traffic flow prediction model,and then introduces the parameters and processing methods of traffic flow.Secondly,it elaborates machine learning and deep learning in detail,laying a theoretical foundation for the establishment of the model.After that,the prediction model based on long short term memory is established,and then the network structure and training process of the model are introduced in detail.After that,the parameters are adjusted continuously to make the model approach the optimal value.Finally,through the comparison with back propagation(BP)neural network model,support vector machine(SVM)model and gated recurrent unit(GRU)neural network,it is testified that mean absolute error(MAE),root mean squared error(RMSE),and mean absolute percentage error(MAPE)of this model are smaller and the prediction accuracy is higher.This thesis also uses the data at different road sections for training.Through comparison,it can be seen that the prediction accuracy of the model for each section selected in this thesis is basically stable at about 93.5%,indicating that the short-term traffic flow prediction model based on LSTM can accurately capture the timing-series characteristics of data and has good stability,which can be applied to the traffic prediction of different sections.