Multi-model system parameter estimation

摘要

Multi-model system is a linearly parameterized system. It is a convenient tool to incorporate supervisory control or higher level discrete-event dynamics in the framework of the linear time-invariant systems. We introduced a parameter estimation problem for multi-model systems. The system is driven by a measured input and a disturbance signal and the output is measured with additive noise. The optimal solution of the multi-model parameter estimation problem is a structured total least squares problem. It is difficult to compute off-line and currently there are no recursive algorithms. We propose a simpler to implement, suboptimal solution and demonstrated by simulation examples its effectiveness. Taking into account prior knowledge improves the convergence of the estimates. We show an estimation procedure with the constraint that the parameter vector belongs to the probability simples. This constraint makes possible to interpret the parameters as probabilities.