Rough Set Attributes Reduction And Its Several Applications In Power System

Rough Set Theory is an extension of Set Theory and a relatively new soft computing tool to analyze and process incomplete information effectively, which has been successfully applied in artificial intelligence, data mining, pattern recognition and other related research areas in recent years. This thesis primarily focused on the study of the core problems associated with Rough Set theory—descretization of continuous-valued attributes and knowledge Reduction, combined with distribution automation, where power transformer fault diagnosis derived from Rough Set theory and Power System Load Forecasting based on Artificial Neural Network have been performed.This thesis introduced the fundamental principles and conceptions of Rough Set systematically, under the framework constructed by which we made several contributions as follows:(1) Discretization of continuous-valued attributes in Rough Set. We introduced two classical methods one of which was discretization based on boolean logic combined with Rough Set and the other is based on importance of attribute. The first algorithm could achieve a more complete discrete division, but with obvious drawback that required a large amount of algebraic computing involved in Rough Set and Set Theory, resulting in ineffectiveness of massive data information processing. On the other hand, the method based on importance of attribute can only approach a discrete division that is non-optimal, despite of the capacity about processing massive data.(2) The problem of attributes reduction in Rough Set. The thesis discussed the classical attribute reduction algorithm based on the discernibility matrix and logical operation, which can achieve a more complete attribute reduction at the expense of high computation complexity that went against the massive data processing of knowledge information systems. And then we introduced an improved algorithm based on importance of attribute and fitness function. This method treated the importance of attribute as heuristic information, accompanied with the participation of Gini index function in the computation of information entropy that employed the fitness as the constraint on the algorithm ending, the benefits derived from which algorithm enabled high computation efficiency due to the avoidance of core calculation of the information decision table, and low computation complexity at theworst case of O(n~2).(3) Electric power transformer fault diagnosis of using Rough Set theory. We proposed a transformer fault diagnosis method based on Rough Set to facilitate the reduction on complex factors, and employed ID3 algorithm to process the reduced attributes and perform the fault diagnosis, which significantly improved the limitation of the decision tree algorithm. Furthermore, combined with Rough Set and artificial neural network, the DGA method rendered a efficient boost of diagnostic accuracy.(4) Power system load forecasting based upon Rough Set theory and artificial neural network. A new method of load forecasting subjecting to incomplete data was introduced, where we employed completing algorithm to refine the original data, and then used the improved reduced discretizaton algorithm to deal with the attributes, followed by the training of attribute through neural network that gave birth to the forecasting results. Simulation results validated this algorithm.