Non-linear Gpc Algorithm Based On Neural Networks

Generalized Predictive Control (GPC) as a new type of predictive control algorithm has great research achievements not only in theory but also in application that are mostly used in linear system domain. But it is difficult to get satisfied control results applying those achievements in nonlinear system domain because it is hard to build an accurate predictive model. Neural Network Control as one of novel intelligent control methodologies can approach arbitrary nonlinear systems according to arbitrary precision that remedy the shortage of inaccurate predictive model. In this paper, we do some research about NN identification and linear GPC algorithm on the ground of introduction to the basic principles of the both. Meanwhile, a nonlinear GPC algorithm based on Neural Network predictive model is presented. In this algorithm, NN is used to build a nonlinear predictive model, and the trained NN parameters are offered to get a linear model through linearising the nonlmear model close to the working point that is presented to GPC. And through simulation the availability of algorithm is be confirmed. In this algorithm, a compensation algorithm is provided to offset the model inaccuracy that is raised by linearization. The algorithm validation is tested by simulation.At last, the quantitive and qualitative theoretical analysis to the choice of the GPC parameters, stability and robustness is given; the methods on the design and choice for the main parameters of GPC systems is discuss; and through a simulation, the correctness of the principle for the parameters choice is proved that further strengthen the study Neural Network nonlinear Generalized Predictive Control.