A computationally efficient method of identifying generic fuzzy models

There is on-going interest in the application of adaptive fuzzy model-based predictive control techniques which attempt to formulate and solve the control problem when the systems are uncertain and non-linear. This paper proposes a computational efficient method to generate a generic fuzzy relational model that can be used to initialize the model used by the controller. The methodology used in this paper is to generate the generic model from computer simulations of a set of different designs by using the “ideal” training data. Methods of reducing the time required to generate the required training data, and transforming a low granularity fuzzy model into a model of higher granularity, are proposed. The generation of a generic model of a cooling coil subsystem of an HVAC system is used as an example to demonstrate how these techniques can successfully identify a generic fuzzy model that is suitable for use in an adaptive fuzzy controller. Results are presented, which show that the time required to identify a high granularity, generic model can be reduced significantly.