Research On Key Technologies Of Traffic Forecast Based On Multivariate Data

Traffic prediction based on multivariate data is playing an increasingly important role in people's daily lives,including traffic flow prediction,network flow prediction,weather forecasting,and so on.It is noteworthy that the traffic flow prediction has become more and more important in recent years.Major companies,such as Tencent,Didi,Baidu,began to develop the products of traffic flow prediction or expanded related functions for their products.It is of great significance to forecast traffic volume based on multivariate data which can provides convenience for relevant government branches,enterprises,and the public.This thesis does research on analysis and prediction of traffic volume and its application.The main contributions of this thesis are listed as follows:(1)An improved SEM-MNL traffic preprocessing model based on multivariate data is proposed.After fully researching the Logit model and the Probit model,the MNL model which is the basic model of the Logit model is selected as the basis of the preprocessing model in this thesis.First,based on the advantages and disadvantages of the MNL model and the SEM,the basic framework of the model is proposed,which is divided into two parts,namely the SEM model of traffic characteristics based on multivariate data and the MNL model considering the influence of potential factors.Then do research on the upper part and lower part separately.In particular,Aiming at the problem of lack of consideration of potential factors,the traditional MNL model is improved by the effect maximization and random utility theory.And the MNL model considering the influence of potential factors is proposed.Finally,an improved SEM-MNL traffic preprocessing model based on multivariate data is obtained after integrating the upper part and lower part.(2)A traffic prediction algorithm model based on sparse partial correlation graph(SPCG-EMD)is proposed.This research is divided into two parts to expand the application scenario of traffic prediction with few multivariate data or abnormal data,namely EMD and SPCG.In order to get a good prediction results,do EMD decomposition of data before forecasting firstly.Then considering the advantages of sparse graphs,reconstruct sparse partial correlation graphs on the basis of partial correlation graphs,and an improved traffic prediction algorithm for sparse partialcorrelation graphs is proposed.Finally,the SPCG-EMD algorithm mode is proposed after integrating EMD and SPCG.(3)A traffic volume forecasting system based on multivariate data is designed and implemented.According to the above research,two models proposed above are both applied to real highway traffic valume prediction.And a traffic volume forecasting system system,which implements to visualize the prediction results,is designed and implemented.Both the preprocessing model and the traffic prediction algorithm model proposed in this thesis have been verified by the analysis of their respective examples,and they are actually applied to the big data analysis platform of the Sichuan highways.