Fuzzy trace identification algorithms for non-stationary systems

In this paper, we propose fuzzy trace identification algorithms for identifying non-stationary stochastic systems. These algorithms are obtained by combining an adaptive trace identification algorithm with a fuzzy logic based supervisor. The supervision level uses the global parametric distance and the signal to noise ratio as inputs. A third input equal to the ratio between short term and long term estimated values of the output prediction error variance can also be used in order to provide faster convergence and better robustness of the parameter estimator in presence of model changes. The efficiency of the proposed identification methods is illustrated by means of simulation examples.