Fuzzy Cognitive Map Approach to Process Control Systems Chrysostomos

Corrventional control has signhicantly contributed to the solution of many control problems, but its contribution to solutions of increasingly complex dynamical systems has practical dWiculties. Requirements in control and in supervisory control cannot be met with existing conventional control theory and new methods are required that exploit past experience, can learn, and provide failure detection and identification. Soft computing thus becomes an important alternative to corrventional control. Fuzzy cognitive map (FCM) usage for control and modeling systems is expected to contribute much to the effort to create more intedigent control systems. FCM describes and models a system symbolically, using concepts to illustrate ditiferent aspects of system behavior that interact, showing system dynamics. A FCM integrates experienoe and knowledge on system operation due to how it is constructed, i.e., using human experts that know system operation and its behavior in ditiferent circumstances. Due to their dynamic nature, FCMs are exploited to represent and conduct system control. Political scientist R. Axelrodi) introduced cognitive maps for representing social scientific knowledge and describing methods used for decision making in social and political systems. B. Kosko6'" enhanced cognitive maps considering fuzzy values for conoepts of the cognitive map and fUzzy degrees of interrelationships between conoepts. After this pioneering work, FCMs attracted the attention of scientists in many fields and have been used in durerent scientific problems. New FCMs have been proposed such as the extended FCM5) and the neural cognitive maps9). FCMs have been used for planning and making decisions in international relations and political developmentsM and have been proposed for generic decision analysis20) and disuibuted cooperative agents2i), FcMs have been used to analyze electrical circuitsi4) and to construct vimal worlds2). ln control themes, FCMs have been used to model and support plant contro14), represent failure models and effects analysis for a system modeliiNi2), and to model the control system supervisori5-i6). The objective of this paper is to define and constmet FCMs models for describing complex systems. Section 2 describes Fems and proposes a calculation rule. Section 3 proposes a soft computing methodology for constructing and developing FCMs. Section 4 implements FCM to model and control a chemical process. Section 5 suggest the use of two-level FCMs to conduct supervisory control and discusses the failure part of a supervisor-FCM. Section 6 gives conclusions and prospects.