Adaptive Neural Network Based Predictive Control of Nonlinear Systems With Slow Dynamics

In this paper a data-driven approach for model-free control of nonlinear systems with slow dynamics is proposed. The system behavior is described using a local model respectively a neural network. The network is updated online based on a Kalman filter. By predicting the system behavior two control approaches are discussed. One is obtained by calculating a control input from the one step ahead prediction equation using least squares, the other is obtained by solving a standard linear model predictive control problem. The approaches are tested on a constrained nonlinear MIMO system with slow dynamics.