Analysis Of Traffic Flow Mechanism Based On Data Mining

For its huge capacity and high efficiency, urban expresswayprovides rapid and cost-efficientservice for traffic users with great social and economic benefit.Currently, our scholarshave gained some achievement in traffic flow theory of urban expressway, but most of themprefer to use math model rather than empirical analysis.For this reason, paper analyses themechanism of urban expressway based onempirical data and data mining, sets up an improvedmodel of traffic phase recognition and prediction using statistics, PCA and Neural Network.Firstly, the paper uses sequence diagram, fundamental diagram and spatiotemporalreconstruction to analyse the data characteristics of traffic phase, phase transition andcongestion propogation qualitatively. Then based on occupancy, the periodic change of varianceof Q、V、htand corresponding time-differential component is used to set up the traffic phaserecognition diagram, and the probability curve of F-S and S-J phase transition is built. Lastly，paper uses PCA and KNN to analyse the traffic data in lower dimension space and choose (Q、V、Occ、ht) as key indicators, which are employed to set up the complementary model of trafficphase recognition and predicition using PCA and Neural Network。The mainly innovation of paper is as follow:（1） Analyse urban express traffic data, set up a traffic phase recognition diagram and fit theprobability curve of traffic phase transition using statistical approach.（2） Use PCA and KNN to choose suitable parameters for PCA features extraction. Set upan improved model of traffic phase recognition based on PCA. Verified by empirical data, themodel works precisely and presents the result with “phase match curve”. This curve shows thegame among three traffic phase and works with NN to predict the traffic phase transition.Research involved in paper such as mechanism analysis and the model of traffic phaserecognition and prediction can provide certain basis to traffic flow theory and ITS controlstrategy.