A new approach to system identification and parameter tuning with multiple model adaptive estimators

Multiple model adaptive estimators have been used as a standard technique in system identification and systems state estimation. In its classical application, this method picks up the most probable model from a fixed set of pre-established models. This paper presents a new method, relying on multiple model adaptive estimators, that defines a set of models, with cardinality dependent on the number of unknown system parameters, and a strategy to successively update the parameters´ values to approximate the system under identification. The method starts with a model for the unknown system and a suitable range of values for each unknown parameter, that can be chosen a priori or generated automatically, and from there a constellation of suitable models is constructed. In a first phase, a search is made in the parameters´ space to seek the region where the true model´s parameters dwell, and after that, the region´s area is consecutively shrunk and the parameters are updated. Due to the adaptive nature of the constellation, for time-variant systems, when the system´s parameters change the algorithm parameters can be reinitialized to estimate the new systems parameters.