On-line identification of multivariable processes using EKF learning-based adaptive neural networks

This paper presents online identification of multivariable processes with time-varying and nonlinear behaviors using two adaptive learning approaches for radial basis function (RBF) neural networks. These approaches are called as growing and pruning algorithm for radial basis function (GAP-RBF) and minimal recourse allocation network (MRAN). The extended kalman filter (EKF) is proposed as learning algorithm to adapt the parameters of multi-input, multi-output (MIMO) RBF neural network in both GAP-RBF and MRAN approaches. Some desired modifications on the growing and pruning criteria in the original GAP-RBF have been proposed to make it more adequate in online identification. The performances of the algorithms are evaluated on a highly nonlinear and time-varying CSTR benchmark problem for comparison purposes. Simulation results show the better performance of the modified GAP-RBF (MGAP-RBF) neural network with respect to the original GAP-RBF and MRAN algorithms.