A multi-rate decentralized parameter estimation method for stiff systems

Parameter estimation methods often assume that no a priori knowledge of the plant is available. Often, as in the case of a stiff plant, qualitative or heuristic information is readily inferred. This paper presents a decentralized method of parameter estimation for stiff systems that utilizes available knowledge of the relationships among the response modes of this type of plant, in particular, the separation between slow and fast modes. The method employs moving average filters to prefilter the input-output data and a system of parallel least squares estimators. The convergence properties of the method are studied and two simulation examples are provided. The decentralized algorithm posesses better robustness properties and imposes a reduced computational load, when compared to a standard least squares approach. Advantages of the decentralized estimator suit it best for application in a noisy environment where standard least squares techniques are known to experience difficulties with stiff plants.