Nonlinear Adaptive Direct Generalized Predictive Control Based on LS-SVR Algorithm

Learning the multi-step forecast optimization strategy from Dynamic Matrix Control (DMC) and Model Algorithmic Control (MAC), Generalized Predictive Control (GPC) has a strong ability to overcome load disturbance, random noise and delay change, and the selected model has less parameter, so it is easy to control. But it also has some problems such as big amount of calculation and no consideration to both rapidity and over modulation. So an adaptive direct GPC method is proposed based on LS-SVR and tracking error. This method uses LS-SVR method to design predictive controller, and uses a modified projection algorithm, based on tracking error, to adjust the weight of LS-SVR adaptively, in order to avoid calculating the inverse matrix. The result indicates that the method not only is very effective but also reduces the calculation amount.