Traffic Flow Prediction Of Short-term Based On CEEMD-RFR And Enhanced KNN

In modern times,the increase rate of vehicles is much greater than the mileage of newly built roads.As a result,a series of traffic congestion problems have emerged one after another.Real-time and accurate traffic flow is the key technology for achieving intelligent traffic control and to ease traffic congestion,but also the objective requirement of establishing intelligent city.Due to the randomness,non-linearity,and time-varying characteristics of traffic flow,the traditional prediction methods based on accurate mathematical models have certain limitations,so machine learning has been widely concerned.As an intelligent model based on statistical learning theory,Random Forest(RF)has the advantages of strong generalization ability,high stability,effective solving of nonlinear and high dimensional problems,and so on.It has some advantages in the research of complex nonlinear science and artificial intelligence.In this paper,the current situation of traffic flow forecasting models is studied,and the advantages and disadvantages of the models as well as the correlation of traffic flow data and the main factors affecting traffic flow changes are analyzed.Thus,an improved model of KNN-CEEMD-RFR is proposed based on the selection of KNN through monitoring station,which combines amalgamation,complementation and decompose of empirical mode.This article systematically expounds the relevant methods of current traffic flow forecasting and points out its shortcomings.It explains the theory of machine learning algorithms in detail and lays the foundation for subsequent forecasting models.A prediction model of CEEMD-RFR based on improved KNN is constructed to predict short-term traffic flow,and the structure,training process,parameter determination and specific steps of the model are explained.Finally,the traffic data of Zhengzhou city road network in Henan Province in Baidu traffic cloud platform was used for verification analysis.Through experimental simulation analysis,the forecasting method of CEEMD-RFR based on improved KNN can effectively filter out the monitoring sites related to the points to be measured and optimize the training set.The trend component and random component of traffic data are separated and compared with the general RFR and SVR models,the errors were reduced by 5.51% and 13.414% respectively,the prediction accuracy was 91%.The results show that the CEEMD-RFR model based on KNN can predict the short-term traffic flow accurately,and its generalization ability and anti-jamming ability are strong.The model is superior not only in accuracy,but also in effectiveness and ease of use.