A Genetic Fuzzy Knowledge Acquisition Strategy Based on the Michigan Approach

• Main Authors: Yeong-Chyi Lee, 李詠騏
• Other Authors: Tzung-Pei Hong
• Format: Others
• Language: en_US
• Published: 2001
• Online Access: http://ndltd.ncl.edu.tw/handle/07455806266123995516

碩士 === 義守大學 === 資訊工程學系 === 89 === Developing an effective expert system requires constructing a complete, consistent, and unambiguous knowledge base. Conventional knowledge-acquisition approaches spend much interview time in building a knowledge base even though many tools have been developed to help with the acquisition process. In this thesis, we thus consider two kinds of knowledge acquisition problems — fuzzy knowledge integration problem and fuzzy knowledge discovery problem. In the first part of this thesis, we consider the fuzzy knowledge integration problem. Since knowledge required to develop a knowledge-based system is often distributed among groups of experts rather than being available from a single expert, integrating the knowledge in different sources is thus very time-consuming. We thus propose a fuzzy knowledge integration algorithm to generate a concise and accurate fuzzy knowledge base from different knowledge sources. The fuzzy rules from different sources are first collected to form a rule pool. These rules are then evaluated by three criteria including accuracy, utility and coverage. Three evaluation procedures, each for a criterion, are thus proposed. An integration algorithm and a set of test objects are used to select good rules to form the resulting knowledge base. The proposed algorithm can remove redundancy, subsumption and contradiction among rules. A concise and compact fuzzy rule base is thus constructed effectively without human expert intervention and thus save much time for knowledge integration. In the second part of this thesis, we consider the fuzzy knowledge discovery problem. A novel genetic fuzzy-rule learning algorithm based on the Michigan approach to automatically construct a fuzzy knowledge base is proposed. The proposed approach consists of three phases: fuzzy-rule generating, fuzzy-rule encoding and fuzzy-rule evolution. In the fuzzy-rule generating phase, a number of fuzzy rules are randomly generated. In the fuzzy-rule encoding phase, all the rules generated are translated into fixed-length bit strings to form an initial population. In the fuzzy-rule evolution phase, genetic operations and credit assignment are applied at the rule level. This phase chooses good individuals in the population for mating, gradually creating better offspring fuzzy rules. The evolution process is iteratively executed until a predefined number of generations is reached. The fuzzy rules in the last generation are then gathered together to form the resulting fuzzy rule base.