Iterative reweighted least squares approach to identification of switched affine systems

This paper proposes a fast identification algorithm for discrete-time switched autoregressive exogenous (SARX) systems. The objective of identifying SARX systems is to identify the switching sequence and the system parameters of all submodels simultaneously. This identification problem is formulated as the mixed integer quadratic programming, which is generally NP hard and difficult to solve. In this paper, we formulate the identification problem as the problem of minimizing a logarithm function and provide a fast least squares based algorithm to obtain an approximate solution. Although the proposed algorithm is heuristic, the numerical examples show that it can identify the switching sequence and the system parameters well and is robust for large noise. This paper also consider the application to the texture identification problem, which can be formulated as the identification problem of 2D SARX systems. Numerical results indicate that the proposed algorithm identifies textures in the image.