Short-term Traffic Flow Prediction Based On Improved Wolf Pack Algorithm And BP Recurrent Neural Network

With the rapid growth of the economic level,transportation has become more and more important to people’s life and the development of the whole city.As people’s travel needs is increasing,their dependence on transportation has also increased,and the city’s motor vehicle possession has reached saturation,which leads to urban traffic congestion.This will paralyze the urban traffic network and cause air pollution seriously.In view of the above problems,the short-term traffic flow prediction in intelligent transportation systems can efficiently predict the traffic flow in the outlet network and provide real-time traffic information for travelers,thus alleviating traffic congestion.Based on the study of short-term traffic flow prediction and swarm intelligence optimization algorithm at home and abroad,this paper proposed a combined model(IWPA-BPRNN)which combined improved wolf pack algorithm with BP recurrent neural network.It aimed to improve the convergence rate of the algorithm and reduce the error of traffic flow prediction.Firstly,this paper chose the real traffic flow data,collected by the transportation research laboratory of the University of Minnesota,as the primary data.Through comparison in the early stage of the experiment,a set of data with better quality and more characteristic of traffic flow data was selected.The data were preprocessed,including addition processing,supplement of abnormal data and normalization of positive and negative data,so that the data can be sorted out as the data form that this paper wants.In the improved wolf pack algorithm(IWPA),because the artificial wolf is easy to fall into the local optimal position in the process of optimization,in order to solve this problem,the parameters which affecting the wolf’s walking behavior were appropriate adjusted,and the running and siege steps of the wolf in the intelligent predatory behavior were improved,so as to achieve faster convergence in the early stage of the algorithm.The structure of the wolf pack algorithm is more complicated.In the later iteration period,there may be a problem that the convergence speed is significantly reduced or even the optimal solution can’t be found.So,this paper selected the improved wolf pack algorithm and BP recurrent neural network model to complement each other.BP neural network has the characteristic of error back propagation optimization,and recurrent neural network has the advantage of fully premeditating the relation between data and data.In the early stage of the prediction,the improved wolf pack algorithm is used for high efficiency optimization,and in the later stage,the error back propagation of is used for optimization.This can effectively solve various problems encountered by the wolf pack algorithm in the later stage and improve prediction accuracy.Finally,this paper completed the implementation of the model with the help of MATLAB platform and the related code,and applied the selected data to test the model’s predicted performance.By comparing the perfonnance of the IWPA-BPRNN model with BPRNN model,it was concluded that the IWPA-BPRNN model had better prediction performance,fast convergence speed and high prediction accuracy.