Identification of multiple mode models via Distributed Particle Swarm Optimization

This paper considers the identification of multiple-mode systems, and introduces a new method to estimate the subsystem parameters of piece-wise affine systems. First, the notion of multiplemode linear regression model and the way to reduce its identification problem to an optimization one are introduced. Second, since the introduced optimization problem is inherently ill-conditioned and nonconvex, a new technique named distributed PSO (particle swarm optimization) is developed to avoid being trapped in suboptimal solutions. The proposed identification scheme can handle the identification of piece-wise affine systems without any prior knowledge about their mode transitions and has no difficulty to handle a large number of data samples, which is an distinguished feature of the proposed method. Finally, an experiment with a set of I/O data from a DC motor system is given to demonstrate the effectiveness of the proposed identification method and to evaluate the performance of the proposed optimization technique.