Data-driven Direct Adaptive Model Based Predictive Control

This paper is concerned with the design of Direct Adaptive Model Based Predictive Control (DAMBPC) using subspace identification technique to identify and implement the controller parameters. The direct identification of controller parameters reduces the design effort and computation load which is usually involved with classical adaptive control techniques. The proposed method requires a single QRdecomposition for obtaining controller parameters directly from input-output data when the model dynamic changes. The method uses receding horizon approaches to collect data and identify the controller. The paper presents a comparison of performance given by proposed control scheme when applied to a 4tank nonlinear system with that of a linear model predictive control scheme and multi-loop PID controllers.