Research On RBF Neural Network Based On KNN-DPC Traffic Congestion Prediction System

With the development of social economy,the living standard of residents has been significantly improved,the vehicle ownership has been increasing year by year.There is an imbalance between traffic demand and supply,result in traffic congestion.The key to the management of traffic congestion is to give warning signal in advance and take preventive measures when the road is congested.Therefore,forecasting traffic congestion has become a main concern in the society.As one of the conditions that affect traffic congestion,environmental factors have a great influence on traffic.None of the data sets in the current study of traffic congestion prediction contain quantitative environmental factors.At the same time,the RBF neural network has been applied to the prediction of traffic congestion in the current stage of research and achieved certain effect.However,in the traditional RBF neural network,the K-Means algorithm is used to determine the basis function center of the hidden layer,which is too dependent on the K value,unable to automatically identify the cluster center and sensitive to noise data.This model will no longer apply if the environmental factors are added to the data set.In order to solve these problems,this paper attempts to use Beijing traffic network data and Beijing meteorological data as basic data,integrating the KNN-DPC algorithm based on normal distribution into RBF neural network,and proposes a RBF neural network based on KNN-DPC model.According to this model,a traffic congestion prediction system is designed,which has high efficiency and accuracy.Research includes:1.In view of the fact that current studies do not quantitatively consider environmental factors in the prediction of traffic congestion,a data set of traffic congestion prediction incorporating environmental factors is designed,by referring to relevant materials and literature,as well as filtering environmental factors related to traffic congestion.2.For the KNN-DPC algorithm,the cluster center is manually selected through the decision graph,which is subjective.Therefore,this paper proposes a KNN-DPC algorithm based on normal distribution,that can autonomously identify the cluster center in the data set,providing the foundation for optimizing RBF neural network.3.Comparing to traditional RBF neural network algorithm that using K-Means algorithm to select the clustering center,this paper uses the KNN-DPC algorithm based on normal distribution as the choice algorithm of RBF neural network basis function,and then obtains the rest parameters through correlative calculations.The optimized neural network model has high prediction accuracy and strong robustness to noise data.4.In order to predict traffic congestion,this paper designs a RBF neural network based on KNN-DPC traffic congestion forecasting system.The RBF neural network is trained by using data sets incorporating environmental factors,which then used to predict road speed,and determine the status of traffic congestion according to that road speed threshold.Finally,in order to verify the feasibility,accuracy and effectiveness of the proposed algorithm,model and system,the KNN-DPC algorithm based on normal distribution is used to perform test of cluster center selection on UCI dataset.The results show that the algorithm has strong feasibility and accuracy.Then the historical road network data and meteorological data of Beijing are used to test the system effectiveness and model accuracy of the traffic congestion prediction system based on KNN-DPC RBF neural network.The result proves that this system can accurately,automatically and effectively complete the prediction of road congestion conditions and has higher prediction accuracy.