Research On Short-term Traffic Flow Forecasting Of Urban Highway Based On Big Data

Traffic jams have become an object of public denunciation in many big cities.ITS(Intelligent Transportation System)is one of the most effective measures to solve them.And short-term traffic flow forecasting and big data are key technologies to realize ITS.How to improve forecasting accuracy and timeliness is the key problem for ITS.Aiming at the technical problems in traffic flow forecasting at present,one year of traffic big data from Twin Cities is used to research key technologies of shortterm forecasting for urban traffic flow such as rapid analysis of big traffic data and multi-model fusion modeling under the big data background.The main innovations are concluded as follows.1.A big data platform Hadoop+Spark for traffic flow forecasting is built.On the basics of researching distributed computing architecture MapReduce of Hadoop and multi round iterative computing system of Spark,an overall framework of analyzing traffic big data and forecasting traffic flow is put forward.Because of the characteristics of large volume of traffic data and strong timeliness of short-term forecasting,a big data platform based on Hadoop+Spark for traffic flow forecasting is built,which combines offline analysis and online forecasting.This platform is applied to mine big historical traffic data with Hadoop/MapReduce and forecast short-term traffic flow with Spark/RDD as well as streaming computing.The establishment of this platform provides a basic guarantee for analyzing big traffic data and forecasting traffic flow.2.MapReduce+JOMP and MapReduce+Fork/Join,two computing models of combining coarse and fine granularity,are established separately.Because big traffic data is mainly composed of a dynamic mass of small files,and frequent I/O operation of massive small files will lead to poor analysis efficiency on Hadoop platform,JOMP and Fork/Join techniques are used to optimize the distributed computing architecture MapReduce.MapReduce+JOMP and MapReduce+Fork/Join,two computing models of combining coarse and fine granularity,are established separately to achieve calculating mechanism of distributed operating between nodes and multi-threads running in parallel within nodes on platform.The experimental results show that two optimizations can effectively improve analysis efficiency of big traffic data.And they are effective explorations for the rapid analysis of big traffic data.3.A combinatorial traffic flow forecasting model based on time series multifractal characteristics is built.On the basis of multidimensional and multilevel analysis of traffic flow spatiotemporal characteristics,combined with intelligent computing,time series and non-linear statistical theory,the combinatorial forecasting idea based on multi-feature decomposition and piecewise modeling according to the data distribution is proposed.And a traffic flow forecasting model based on time series multifractal characteristics is built.Experimental results show the proposed model can effectively improve forecasting accuracy of short-term traffic flow and the forecasting results of multistep ahead is more better.At the same time,it is proved that combinatorial forecasting method which takes into account linear and nonlinear multimode information of traffic flow and integrates traffic flow characteristics analysis and forecasting is an effective method to improve forecasting accuracy.4.A combinatorial traffic flow forecasting model based on wavelet analysis and multivariate time series is built.The combinatorial traffic flow forecasting method based on multi-source traffic data is studied.Based on wavelet analysis theory and multivariate time series modeling,WSARIMAX,a combinatorial model for short-term traffic flow forecasting,is built,which is based on wavelet analysis and multivariate time series.So the modeling limitation that multivariate series must satisfy cointegration relationship,is expanded.Experimental results show the proposed model can effectively improve forecasting accuracy of short-term traffic flow.It is concluded that under current traffic detecting level the influence of data quality on forecasting accuracy is less than that of external factors,which indicates that multi-source big traffic data plays an important role in improving the forecasting accuracy of short-term traffic flow.This research focuses on solving three technical problems of rapid mining of big traffic data,deep analysis of spatiotemporal characteristics of traffic flow and multimodel fusion modeling of traffic flow forecasting which has important significance for innovative research on short-term traffic flow forecasting under the big data background.