Research And Application On Rough Set-based Adaptive Neuro Fuzzy Inference System

As a new method of knowledge discovery and data analysis, rough set is becoming the foundation theory of artifical intelligence and recognize science. It can handle unaccurate and incomplete information, and draw implicit knowledge from this information. Related to other theory tools of uncertainty and vagueness, rough set has many unreplace superiority.ANFIS (Adaptive Neuro-Fuzzy Inference System) is the most successful system which is the combination of fuzzy logic and neuro network, the adaptability of ANFIS make it can replace any neuro network and conduct same function. However, this kind of network has three problems: large calculation quantity, low intelligent degree(rely on field expert knowlege),lake of the function that handle indetermined information,The major method improving ANFIS concentrate on: simplify network structure,reduce the complex degree of network.Inspire information entropy and rough set theory are concerned in the development process of ANFIS it is rare now. Through analysing the three defects existed in ANFIS, comparing the characteristics of rough set and the concept of information entropy and pursuiting the theory foundation of their combination. It will be find that applying the information entropy and rough set theory is hopeful to resolve the problems of ANFIS. It can perfect the network structure and enhance it's intelligence. We can estimate that the combination of them will promote intelligent control development further.Commencing with pursuiting the possibility and necessariness the combination of ANFIS and rough set, and seeking the point of the penetration of both, this paper aims at constructing more perfected rough set-based ANFIS system and apply it in actual control system.The specific thought and method of constructing system:Contruct fuzzy-neuro network and pretreat network data appling with roufh set. This paper present the thought that build therule of network through pretreating network data and extracting rule from actual example, and a new method of attribute reduction and rule extraction is put forward.The key to affect actual control effection is selecting the input variable reasonabily, this paper attempt resolve this problem by using the concepe of information entropy in rough set. Selecting the parameter with large relativity as input. It can be assure the reasonability of parameter fully, reduce the calculate workload of system, save converge time of network, and enhance the forecast precision of model at the same time.Rough set-based ANFIS system need not to carry out structure optimization, so it is simple to be realized by optimizing former and rear parameters using hybrid algorithm. The difference between structured system and ANFIS is the combination with input layer and membership function layer is not completely combined-type connection again, so network structure is simplified greatly.Actual control experiments show that, this strategy solves rule explosive problem and avoids difficulty of parameters option and drawback of designing membership function subjectively, It open awide universe for the development of intelligent control.