A Research On Rules Extraction By Method Of Data Mining Based On Cloud Model

Extracting rules from Data Base (DB) is an effective way of constructing knowledge model of a control plant. However, when applying this method, we have to handle a problem, i.e., how to divide the input and output data space effectively and properly. Currently, some problems such as solid partition, intelligibility of rule, approach performance of rules, completeness and robustness of rule base have not been dealt with in existing rule extracting methods. In this paper, we will use Cloud Model as research measure, and propose a new method—Scale Cloud Transform Method (SCTM), and also do some valuable work about those above problems.Mainly, our research try to deal with the most important problem in extracting rules—how to divide the data space, including the input and output data space. We will introduce Cloud Model into our research, and make use of its merits in transforming between the qualitative and the quantitative value to discover the new method of extracting rules. For the problem of solid partition, Key Point (KP) is introduced into our new method, and we get KP by statistic calculation, not by expert knowledge or rules, so rules got by our new method can embody the information of data distribution, and easily to be understood. Furthermore, the Best Divided Point (BDP) is introduced to make sure that the partition of data space is dynamic and changeable, so we can adjust the partition of data space according to approach performance. BDP also help to make a complete rules base. About the completeness of rales base, we propose the definition of Interval Approach Error (IAE). Our new method firstly process data and generate rules interval by interval, and then calculate IAE and compare it with the general error, according to the result we adjust the rules base, including dividing an interval and combine two intervals.For validating and proving the validity and feasibility of our new method, we study and explore the merits of T-S fuzzy controller, and then we propose T-S Cloud Model Controller. Then we design a simulation experimentation based on the new controller. The result show that: SCTM can divide the data space properly and effectively, and the partition can embody the distribution of the data space. Beside smaller error, SCTM can reduce the number of rules extracted from sample data, and can make rules well be understood; the rules base of the new T-S cloud model controller is complete and robust, and easy to be applied.Our research introduces Cloud Model into the domain of rule generation and optimization, and proposes a new method for generating rules. Furthermore, as for the research and development of Cloud Model Theory, our research also has certain academic value.