An adaptive multi-controller architecture using particle filtering

We propose a method for the supervision of a multi-controller robust adaptive control architecture based on particle filtering, and apply it to control a two-input two-output plant characterised by parametric uncertainty and potential time-variability of the uncertain parameters. The state variables are augmented to include the time evolution of the uncertain parameters; moreover, the probability distribution of the state estimate given by the particle filter is used to weight the control action of a bank of robustly designed controllers. Monte-Carlo simulations are used to compare the performance of the new approach with that of a robust non-adaptive controller, an extended Kalman filter approach, and a hypothetical system capable of perfect plant identification. Results indicate that the particle filter supervisor delivers comparable performance with other approaches both in the presence of constant uncertain parameters and fast parameter variations.