Research On The Traffic Flow Forecasting Method Based On Neural Network In Hadoop Environment

With the rapid development of economy,transportation has rapidly become the backbone of national economic development.Although the rapid development of the transportation industry has brought great convenience to people,but the ensuing serious traffic congestion problems.Real time and accurate traffic flow forecasting is the key technology in the Traffic Guidance System,and the Traffic Guidance System is an important component of the Intelligent Transportation System.Because the traffic system is a human factors involved in the non stochastic steady system,more and more traditional linear prediction model is not suitable for nonlinear traffic prediction,intelligent prediction and optimization model has attracted more and more attention.In this thesis,the representative forecasting methods that accord with traffic flow data characteristics are deeply studied.The classical method of artificial intelligence is analyzed and BP Neural Network is used as the basic algorithm of traffic flow prediction.In traditional BP Neural Network,the training time and training accuracy can not be guaranteed at the same time when traffic flow forecasting is carried out.Firstly,this thesis analyzes the predictive model of the traditional BP Neural Network,and proposes a prediction model to accelerate the convergence of the dynamic value of the output layer to the fixed value.Secondly,on the basis of the fixed output layer,aiming at the shortcoming of long training time of neural network,a K-BP prediction model using K Nearest Neighbor algorithm to optimize training data set is proposed.In the model,the training data set of BP Neural Network is selected before considering the matching degree between predicted data and training data.Compared with the traditional neural network,the training time and training error of this model have been improved.With the extensive application of information technology and Internet of things technology in urban transportation field,the data of urban traffic flow has shown many characteristics of big data.The traditional neural network prediction model can meet the demand of traffic flow prediction under the premise of small training samples.However,with the increasing size and data of training samples,the traditional neural networks often consume too long time in training samples,which is not conducive to the realization of real-time short-term traffic forecasting.This thesis proposes the prediction model of distributed processing framework and combined with BP Neural Network MapReduce in the Hadoop environment,the use of BP Neural Network model of MapReduce parallel and prediction accuracy in which reduce the prediction time,achieve the real-time prediction.