The Modeling And Control Of Dynamic System Based On Fuzzy Cognitive Maps

As a simple intuitive graphical representation and efficient inference mechanism, fuzzy cognitive map has been widely used in many fields, such as medical domain, in-dustrial process control, and environmental monitoring. Fuzzy cognitive map is the com-bination of fuzzy logic and neural network, which is fit for the description, prediction and control of nonlinear dynamic systems. Due to the limitation of human experience, knowledge and cognitive ability, it is very difficult for the experts to construct the fuzzy cognitive map of large-scale intelligent systems in many fields. Recently, more and more researches on the automatic or semi-automatic construction of fuzzy cognitive map based on the dynamic data have attracted a lot of attention of scientists. This dissertation firstly summarizes, compares and analyzes the construction, weights learning, stability analysis and practical applications of fuzzy cognitive map. Then it describes the research process of fuzzy cognitive maps construction, weights learning and control based on dynamic data. The main topics include:(1) After the detailed analysis of the drawbacks of the conventional fuzzy cognitive map transformation function, the dissertation presents an improved method which draws the coordinate coefficients into the fuzzy cognitive map transformation functions. Without a forehand specification, these coordinate coefficients can be obtained from the system data by means of automatically learning algorithms. The improved function makes the fuzzy cognitive map to be used to describe reality systems more accurately.(2) It is known that the learning methods of the fuzzy cognitive map weights arc mainly divided into three types which are Hebbian technology, genetic algorithms and swarm intelligence. These three fuzzy cognitive map weights learning algorithms all need to be iterativelv computed. So they all have overweight computational load. The dis-sertation combines the method of least squares with the learning of fuzzy cognitive map weights, puts forward a novel, convenient, simple, accurate and more rapid approach to the fuzzy cognitive map weights learning based on least squares. Using this approach, the only work needed to do is to solving simple linear equations to get fuzzy cognitive map weights, not needed to undertake the iterative calculation. Thus, this approach can greatly improve the efficiency of the learning algorithm. It needn't specify any coefficient or initial value ahead. And it is an unsupervised learning algorithm.(3) Aimed at large nonlinear systems. this dissertation proposes a split fusion ap-proach which divides the whole state space into numerous smaller local areas in which a fuzzy cognitive map is built respectively. The fuzzy cognitive map with a real-time connection matrix is built on the system state by T-S fuzzy model. Using this approach, we can solve the describing problem of large, complex systems which can not be described in a fuzzy cognitive map.(4) The dissertation does also describe an automatic control algorithm of dynamic system through the combination of fuzzy cognitive map and Hebbian theory, which can be executed step by step and define the precision by itself. The method controls system gradually approaching to the target through the separation system control goals until it finally meets the control accuracy.In the last, the dissertation proves the validity of fuzzy cognitive map modeling and control theory through the complex nonlinear truck back-upper example. The experi-ment results in the dissertation indicate that these approaches are practical, accurate and efficient.