Parameter estimation of dynamic fuzzy models from uncertain data streams

Modeling of time-varying dynamic systems in real time requires the use of streams of sensor data and incremental learning algorithms. This paper introduces an incremental fuzzy modeling approach based on uncertain data streams. By uncertain data we mean data originated from unreliable sensors, imprecise perception, or description of the value of a variable represented as a fuzzy interval. An online incremental learning algorithm is used to develop the antecedent part of functional fuzzy rules and the rule base that assembles the model. A recursive least squares-like algorithm updates the parameters of a discrete state-space representation of the fuzzy rule consequents. Data uncertainty is accounted for using specificity measures of the input data. An illustrative example concerning the Lorenz attractor is given.