Study On Attribute Redution Based On Rough Sets And Its Application

Rough Sets theory is an effective tool to analyze and deal with the incompletion and imprecision knowledge, and it has become a hot research point in intelligent information processing. Currently, the Rough Sets theory has already been applied extensively in the field of Machine Learning, Pattern Recognition, Knowledge Discovery, Decision Analyzing and Procedure Controlling etc. In this dissertation, the algorithm of attribute reduction based on Rough Sets is discussed, and a new algorithm of continuous attribute discretization and an algorithm of attribute reduction based on attribute similarity are proposed, finally this new algorithms are applied to petroleum well logging for oil layer identification. The main works are as follows:(1) The typical discretization algorithms are analyzed, and the shortages of which are pointed out, namely there is no uniform theory to select the candidate cut-points, moreover, the candidate cut-points are not reasonable or the number of them is too large. For this reason, a new algorithm of continuous attribute discretization based on the characteristic of the curve inflection point is presented. In this algorithm, the curve inflection points are obtained to be candidate cut-points firstly according to the subsection character of the curve reflected by the practical data, then the Particle Swarm Optimization (PSO) algorithm is used to optimize the set of candidate cut-points. The experiment shows that the algorithm is feasible and effective.(2) Current reduction algorithms are discussed and there are some problems in attribute reduction based on attribute significance, such as the reduced results are different with the different definitions of attribute significance on the same sample, the minimal reduction set is hard to be obtained when several attributes possess same significance value, and so on. Therefore, the concept of attribute similarity is given according to the similar relation between condition attribute and decision attribute in this paper, moreover, an attribute reduction algorithm based on attribute similarity is presented, namely the condition attribute of which the similarity value is minimal is deleted. This algorithm is convenient and practical, and the simulation result shows that the algorithm is feasible and optimal.(3) Attribute reduction about well logging is an effective method to solve the problems in complex oil layer identification, and which can also save cost and provide serves for predication and decision of petroleum. For this reason, an oil layer identification system based on attribute reduction and Least Square Support Vector Machines (LS-SVM) is presented in this paper, and it is applied to oil layer identification of key wells in a certain oil field. Practical application shows that the result of oil layer identification is accord with that of oil trial, and the application effect is notable.