Direct adaptive predictive control of an hydro-electric plant

Even if predictive control has proved to be an efficient control strategy, the maintenance of a high level of performance may be impossible to reach with a fixed predictive controller in case of important parameters variations. A solution is then to develop an adaptive version of the predictive controller for systems with parametric disturbances. The paper presents a direct version of adaptive generalized predictive control, where the algorithm is rewritten in an original form minimizing a performance index. Developments include a least-squares type strategy for the controller parameters online identification and a conditional updating test in the adaptation loop. An application of this structure to an hydro-electric plant with nonlinear variations of the surface is finally developed. The results are compared to performance obtained with a classical PI controller.