Design of Multivariable Self-Organizing Fuzzy Decoupling Controller Based on Neural-Network

To solve problems of multivariable decoupling controller structure complex and difficult to determine, and training times long etc in the absence of a prior knowledge, A new neural work structure - self-organizing fuzzy controller and on-line algorithm - is proposed. The controller is constituted by three-layer RBF neural network. Dynamical structure training based on error criterion and new adding technique can start with zero or single rule, network structure i.e. fuzzy rule is expend and optimized, the method is very simple and the calculating amounts is reduced. The parameters i.e. membership function is self-adjusting by gradient descent. The controller are optimized by on-line updating learning. The proposed control scheme is used with design of multivariable fuzzy decoupling controller for distillation column. Simulation result shows the proposed method is effective, the convergence of trajectory is quicken, the performance of system is obviously improved