New SISO and MISO Adaptive Nonlinear Predictive Controllers based on self organizing RBF neural networks

This paper proposes an adaptive nonlinear model predictive control (NMPC) scheme for control of nonlinear time-varying processes. An online identification approach based on an adaptive modified growing and pruning radial basis function (MGAP-RBF) neural network is presented to model the process dynamics in real-time operations using an unscented Kalman filtering (UKF) estimation algorithm with a forgetting factor scheme. A single-input, single-output (SISO) adaptive NMPC controller is described based on quasi-Newton approach. The adaptive NMPC control procedure is extended to multi-input, single-output (MISO) processes. The proposed SISO and MISO NMPC controllers are evaluated on a highly non-linear time- varying nonisothermal continuous stirred tank reactor (CSTR) benchmark problem. The simulation results demonstrate the potential capabilities of the two developed NMPC controllers to identify and control the CSTR process with superior performance over the conventional PID controller.