Generalized self-organized learning modeling and model predictive control for nonlinear plants

After investigating the difficulty of the existing backpropagation based modeling and inverse control methods, a hierarchic neural network model for nonlinear plants and its learning algorithm-generalized self-organized learning (GSL) is proposed in this paper. In this new model, the state space is quantized by a group of cells, and multiple local models are adopted to characterize the specific plant nonlinearity in each cell respectively. A new model predictive control (GSLMPC) algorithm is also developed accordingly. By exploiting the local model feature of GSL model, the GSLMPC turns the complicated nonlinear optimization into finite iterations of explicit linear solution. In this way, the online computation burden can be reduced significantly. Simulation results show the advantages of the GSL and GSLMPC