Generalized predictive control of nonlinear systems using evolutionary computation

The present study proposes a novel method of designing generalized predictive controller (GPC) for nonlinear systems using evolutionary computation. The success of GPC usually depends on the prediction accuracy of the models which are being used to predict the output. The proposed controller is based on the output predictions from an extended model which consists of a linear controllable model (called the reference model) and a disturbance term. The disturbance component varies with time and accounts for the nonlinear dynamics which can not be modeled by the linear reference model. The success of the proposed GPC depends on the accurate estimation of the disturbance term. The disturbance term is represented by a polynomial NARMAX model using evolutionary computation such that the output of the extended reference model matches with the output of the nonlinear system. The generalized predictive control law is computed based on the extended linear reference model, which includes the effects of nonlinearity, following standard design methods of linear GPC and therefore offers significant computational advantage. Optimum value of some of the tuning parameters of the GPC such as control and prediction horizons are obtained using evolutionary programming. The performance of the proposed method has been illustrated considering several examples of nonlinear system and has been shown to be satisfactory.