A fuzzy fusion algorithm to identify the traffic states by the floating car data

Abstract

Floating car data provided the speed of each section every five minutes. The clustering methods could reflect the fuzzy character of traffic states, but the parameter threshold would be man-made subjectivity somehow. The neural network could solve this problem, but the floating car data could not be used directly because of the training time as the data was too massive. Based on the reality road network of Beijing, this paper divided urban traffic states into four levels which were blockage, congestion, slight congestion and free flow. To overcome the problems above, this research used a fusion algorithm which connected k-means clustering with MLP neural network to identify the traffic states. This method had a high reliability and the result was consistent with the actual traffic condition. This paper finds a fusion algorithm to get the threshold value of different level urban roads by the floating car data which just needs a short training time. This paper also gets the conclusion that the dipartite degree is not inadequate between the urban expressway and the urban main road if the traffic state is classified into three types.