Research On Short-term Traffic Flow Forecasting Method Based On State Discrimination

As urban traffic congestion associated with environmental pollution and traffic safety issues have become increasingly prominent,as an important way to solve the problem of traffic congestion,the research of Intelligent Transportation System(ITS)research increasingly gained more and more attention.The traffic flow prediction technology is not only the core of the intelligent traffic management system,but also is the basis of intelligent traffic control and induced.Accurate traffic prediction information has significance meanings for path selection,saving travel time,alleviate traffic jam,reducing exhaust pollution and saving energy.Currently the existing models and methods used in many fields of control hundreds,but no prediction model is the best in all respects.They both have their own advantages,disadvantages and applicable situations.For example,based on the traditional statistical theory of ARIMA model is simple and easy to understand.And for stationary series has high prediction accuracy in the case of sufficient data.While in the estimated parameter,it has to rely on a large amount of data.Neural network model is particularly suitable for complex transportation,but unfortunately its portability and replicability is poor.Therefore,this paper proposes a new traffic flow forecasting model,which according to the current traffic parameters to analysis related traffic state,and based on status results to select the appropriate forecasting model matched this state.This paper mainly completed the following work:First of all,the view of the traditional FCM clustering algorithm in the lack of traffic state identification,what improvement in this paper is the respect which selects the initial cluster center.Fuzzy C-means algorithm for the initial cluster centers associated with a strong resistance,will easily lead to fall into local optimal solution,so before making FCM clustering,we should choose a suitable initial cluster centers.In the proposed approach,based on maximum density clustering process is carried out first,the clustering center will be set as the initial value of fuzzy c-means algorithm,then get the final cluster centers according to the FCM algorithm.According to the clustering center,compute each pending data points about each state with membership,select state with largest membership degree as a state discriminant result.The experimental results show that the improved FCM algorithm has less misjudgment rate.Secondly,according to the results of state identification,the model of traffic flow forecasting is selected,and improved the model by using the method of constructing the combined forecasting model.Basic method is based on FCM algorithm to judge state,and then select the traffic flow forecasting model to predict according to the state.While in the process of state transition,according to the traditional rules only select one state corresponding forecasting model to forecast there may be a large error.So in this case,according to the value of the membership of two basic choose two forecasting model consisting of a combination of model,namely respectively using two models to forecast,and the results are weighted processing which weight is calculated according to the membership.Experiments show that the combined forecasting model based on traffic discriminant than a single model and based on the state criterion of single model has higher accuracy.