Online identification of switched linear output error models

In this paper, we describe a modified recursive least squares (RLS) algorithm for the online identification of switched linear systems (SLSs). For this problem, a real time mode detection (MD) method is usually employed to detect the running mode for each output data and this part is very important for the estimation results. In fact, it is rather difficult to design a perfect MD method that can identify the running modes without errors, especially in stochastic systems. As a result, there often exist some mode mismatches in the MD procedure. For this reason, we cope with this problem from the compensation point of view. By introducing a resetting strategy to the RLS algorithm, the negative effects of mode mismatches will be separated into a few resetting intervals, which can effectively avoid them to be accumulated and thereby result in good estimation. The performance of the proposed algorithm is evaluated by Monte Carlo simulations in comparison with another alternative method.