Neurofuzzy Model Construction for the Modelling of Non-linear Processes
K.M. Bossley, Michael Brown, Chris J. Harris · ePrints Soton (University of Southampton) · 1995
Neurofuzzy systems are ideal for modelling nonlinear processes; combining the transparent knowledge representation of fuzzy systems with the well established learning techniques of associative memory networks. In the control community there is a need for high-dimensional modelling, but unfortunately these neurofuzzy systems suffer from the curse of dimensionality. To make the use of high-dimensional neurofuzzy systems practical, off-line construction algorithms need to be developed which generate parsimonious models. This need for off-line construction algorithms is enhanced by the lack of expert knowledge from which models are traditionally developed. This paper outlines the concepts of B-spline neurofuzzy systems and presents several representations which can be exploited to produce parsimonious models. Finally, an off-line construction algorithm based on several of these representations is illustrated with a simple example.