Constructive on-line learning for a neuro-fuzzy network with fuzzy sets obtained by Delaunay triangulation

This paper addresses the design and gradual building of a rule based neuro-fuzzy network using piecewise linear multidimensional membership functions obtained by Delaunay partition of the input space. Online growing and pruning techniques are used to obtain a parsimonious structure. The proposed network is shown to be useful in approximating unknown nonlinearities of dynamic systems. A control framework is applied, taking advantage of the piecewise linear property of the model. For each simplex, the local inverse model can easily be calculated. The operation of this adaptive control scheme using the online constructive algorithm and the inverse of the local linear model is demonstrated using a simulation example and a laboratory scale process