Font Size: a A A

Research On Genetic Fuzzy System For Knowledge Extraction

Posted on:2005-07-16Degree:DoctorType:Dissertation
Country:ChinaCandidate:H ZhangFull Text:PDF
GTID:1118360152456684Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
Fuzzy theory, established by Zadeh, has been a powerful tool for describing, manipulating the fuzziness of the things and the uncertainty in the systems and simulating fuzzy logical thinking process specially owned by human beings. Fuzzy logic is a computational paradigm that represents and manipulates information in a way that resembles human communication and reasoning processes. A fuzzy system (FS) is any Fuzzy Logic-based System. Fuzzy Rule-Based System (FRBS) can be divided into two kinds: approximate and descriptive. The former is mainly concerned with accuracy and flexibility. The latter is focused on interpretability. The trade-off between accuracy annd interpretability is an important research topic.The basic task of knowledge extraction is to acquire the knowledge for experts, which is predictive model or descriptive model, so as to build the sound and effective knowledge base for people's need to solve real-world problem. People can't extract knowledge from a large amount of data in many applications, which motivates them to study the computer technology used to extract and represent knowledge in Fuzzy Rule-based Systems (FRBS), it is the nature of data-driven fuzzy modelling. Data-driven fuzzy modelling has been applied to many fields, such as pattern recognition, data mining, classification, prediction, and process control.In the nineties, despite the previous successful history of fuzzy systems, the lack of learning capabilities in the field generated a certain interest for the study of fuzzy systems with added learning capabilities. Two of the most successful approaches to integrate learning capabilities have been the hybridization attempts made in the framework of soft computing, were different techniques, such as neural network and evolutionary algorithm.Evolutionary computation is an effective method to solve optimization, search and learning problems, inspired by genetics and nature evolution. Evolutionary learning process covers different levels from parameter optimization to learning the rule set. Besides that, GA is among of the few powerful tools to extract the knowledge. Therefore, fine tuning techniques and adaptation have been introduced into fuzzy systems, meanwhile, research as well as applications have been gone one step further from pure tuning to actually extracting knowledge from data in terms of fuzzy rules. In the background of fuzzy logic and evolutionary algorithm, this dissertation studies evolutionary modelling and deals with trade-off between accuracy and interpretability in terms of two kinds of fuzzy systems so as to extract fuzzy rule-based models to solve classification and identification problems. The main contributions of this thesis are as follows: Gives a brief overview of the concepts and features of fuzzy sets operators, fuzzy implication, fuzzy rules and the techniques of fuzzy modelling and evolutionary learning, this is the theoretic and practical basis of the study of genetic fuzzy system in theory and practice.Studies the hybridization of fuzzy clustering and genetic algorithm (GA) to solve identification of TSK model. First, fuzzy c-mean clustering algorithm is employed to search for premises of fuzzy rules, weighted recursive least square method to get the consequents, and similarity analysis to reduce the rule set through rule combination and deletion; a genetic algorithm encodes membership function and consequent parameters and search the space of GA to accomplish the parameter learning process. The experiments show the resulting model's high accuracy and simplicity. Studies a Fuzzy Rule-based classification system to extract from complex, high-dimensional data in the framework of singleton fuzzy system. In order to improve the abilities of linguistic fuzzy modelling, this thesis proposes a method of generating the hierarchical rules with different granularity hybridization in multi-stage FRBS learning process. GA is employed to select the rules and get the weights, wherein binary code (for rule selection) and real value code (weight)...
Keywords/Search Tags:Fuzzy Logic, Fuzzy inference, Fuzzy Modelling, Fuzzy Rule-Based System (FRBS), Evolutionary Computation, Fuzzy Clustering, Co-evolution, Interpretability, Pattern Classification, Model Identification.
PDF Full Text Request
Related items