Adaptive identification of Fuzzy Cognitive Networks

An adaptive methodology is introduced to identify the weights of a general Fuzzy Cognitive Network (FCN). The proposed scheme consists of two hierarchically structured individual identification tasks that are both implemented on-line. The inner task employs a robust integral of the sign error (RISE) observer to approximate the system´s unknown nonlinearity. At the higher level, a nonlinear mapping of the observer´s output provides an estimation of the linear parametrization of the FCN weights. Exploiting this fact, a standard gradient-based algorithm is proposed as an outer identification task to estimate the FCN weights under a persistence of excitation condition. Simulation results illustrate the effectiveness of the proposed approach.