Data Mining Based On Cloud Model And Its Application In Traffic Flow System

Data mining has been being a hotspot in academe these years. Cloud model which was put forward by Academician Lideyi is an effective tool for converting quantitative data between qualitative concepts. It has been successfully applied to mass data mining.Traffic command center stores mass data, from which the law and knowledge of the traffic flow can be discovered through data mining. The knowledge can be applied to more effective transportation planning, traffic management, and traffic control. The traffic state identification is an important step in traffic control. Its first step is to fix on the traffic state classification while the traffic state clustering is the premise of the traffic state classification.This paper mainly includes three parts:In the first part, related knowledge of data mining involved in this paper and related concepts of cloud model have been briefly introduced. The normal cloud, normal cloud's mathematical quality and its general applicability are the main contents.In the second part, the concept of concept level and its auto-generation is briefly introduced. An improved cloud transformation algorithm has been put forward, in which threshold value which is set by user during calculate the proper entropy has been get rid of, making the initial generating concept sets more objective. Base on the"3En principle"a concept partition algorithm based on cloud model attribute similarity has been put forward. A more all-sided definition of truncation entropy than the exist for cloud combination operation has been given. A criterion algorithm of concept level partition based on cloud mode has been put forward, which is convenient for user to fix on the concept granularity on a certain concept level and choose proper algorithm of concept partition. Finally determinant method of membership concept is briefly introduced.In the third part, the cloud model is introduced into the traffic data mining, and applied to concept partition of the collected traffic data, confirming its effectivity in the traffic state clustering. Using two groups of collected traffic data, the set threshold value's effect to the generating concept is confirmed and applicability of the two concept partition algorithms, one of which is based on the nearest distance between two expected values, the other is based on the cloud model attribute similarity has been discussed. The results illuminate that choosing algorithm must be according to the data distribution and the concept level granularity.