Model Method And Application Study Of Fuzzy Cognitive Map

Fuzzy cognitive map (FCM) is one of hotspots in soft computing methods. In this dissertation, we analyze the advantages and the drawbacks of FCM on the basis of early research. However, FCM lacks system analysis method and the capability of learning, the advantages of FCM aren't utilized, this dissertation focuses on the analysis of FCM based on graph theory, learning and the knowledge fusion from different human experts, the modeling of complex system and the control applications of FCM are discussed.The main work and conclusion are as follows:1. The structure, formalization and inference mechanism of FCM are discussed, the dynamic causal relationship is introduced, and the causal knowledge representation based on FCM is directly shown with concepts and the relationship of concepts. The dynamic action of system is simulated with the action between concepts in FCM, the carrying out of FCM reference is by recurrence action from fore nodes to back node.2. On the basis of lack system analysis method, we analyze the causal relationship and the causal influence transfer among concepts using the feature of adjacency matrix and accessible matrix of graph theory and give the measurement of causal influence degree. The causal chain algorithm is presented, the algorithm aims searching all causal chain using accessibility of concepts. In addition, the feedback circle algorithm is presented, the algorithm solves the corresponding basic feedback circle of residual tree based on basis circle definition and the relationship between spanning tree and residual tree of spanning tree, then, all basis feedback circles are given by "circle add" of basic feedback circles.3. On the research of complex FCM, the decomposition methods based on connectivity and genetic algorithm are researched. A complex FCM can be taken for composition of many strong connective blocks. The strong connective blocks are equipollence class, it can produce a decomposition of node set V (G). We present a decomposition algorithm of FCM based on strong connectivity, the algorithm decomposes FCM into several basic FCM by cognizing strong connectivity block with accessible matrix. In addition, as the decomposition based on the connectivity is a strict decomposition, it may be not proper to some applications, we also present the decomposition method based on genetic algorithm. The algorithm uses module thought and establishes partition rules according to feature of FCM, it can decompose FCM into sub graphs using optimization method according to the decomposition rules. This research can increase cognition to structure of FCM, and have actual meanings and directive action on the research of inference, complex system modeling, the application of control,4. Learning ability is the base of FCM self-determination behavior. In this research, we divide the automatic construction problem of FCM into two cases according to feature of problem. First, the system provides only sample data. Second, the system provides not only sample data, but provides expert knowledge. We transfer learning of FCM to optimization problem and imitate the self-organization and self-learning mechanism, the approach of FCM learning is presented based on genetic algorithm. In this algorithm, we use DNA coding to represent chromosome for using adjusting and control mechanism of gene. In addition, we present the immune genetic algorithm to learn FCM model by combining biologic evolution and biologic immune. The presented algorithm uses immune concept on the basis of genetic algorithm, the feature of FCM and expert knowledge are used on search process. The core of algorithm is constructing of vaccine and vaccination. The characteristics of system and the experts' knowledge are abstracted to be a schema, the schema is formulated to generate new individuals. The aim of vaccination is to restrain the degenerative phenomena during evolution and increase individual quality. The study results show that the proposed method is capable of automatically generating FCM model. The illustrative example of the stock prediction suggests that the method is efficiently.5. In application of multi-expert construction FCM, in order to avoid individual expert's knowledge subjectivity, one-sidedness and limitations, we study the solving strategy of multi-expert system based on D-S evidence theory. This strategy can gradually reduce the hypothesis sets and approach the truth with the accumulation of evidences, which make the result of decision more all-around and more scientific. In the method, we use multi-expert's knowledge as evidence, the possible value of weight as frame of discernment, expert's evaluation to a weight on frame of discernment as basic probability assignment, and use D-S rule combining to give fusion basic probability assignment m. Finally, the weight is given according to fusion basic probability assignment. The result is shown that the method can keep exactitude information, reduce conflict factor, strong degree opinion and improve knowledge quality6. On engineering applications, we present a FCM-based general control framework and a FCM-based self-adapt control framework both of which combine control theory with FCM theory. The research use the model of FCM feature and the inference mechanism the causal relationship among variables is constructed by manual or learning. The values of control variables are given by the inference of FCM model. Finally, the system uses the control variables to adjust the controlled variables in actual process and carries out the control of multi-input and multi-out control.The dissertation research FCM from different aspects and different levels, the researches and new opinions are made on the basis of early research and future work.The main innovations of this thesis are summarized as following:1. We analyze the causal relationship and the causal influence transfer among concepts using graph theory and give measurement of causal influence degree, the causal chain algorithm and the feedback circle algorithm are presented. On the research of complex FCM, the decomposition methods based on connectivity and genetic algorithm are researched.2. An approach is presented to fusion multi-experts knowledge based on the Dempster-Shafer evidence theory.3. The approach of FCM learning is presented based on genetic algorithm. In this algorithm, we use DNA coding to represent chromosome for using adjusting and control mechanism of gene. In addition, on order to utilize the expert's knowledge and the feature of system, we present the immune genetic algorithm to learn FCM model by combining biologic evolution and biologic immune.4. On application of FCM, the application method of FCM in control area is research. The general control and self-adapt control framework are presented.