An On-Line Learning Algorithm of Parallel Mode for MLPN Models

摘要

An on-line learning algorithm in parallel mode for multi-layer perception network (MLPN) model is proposed. The MLPN is on-line trained directly in a parallel mode. The on-line learning algorithm is based on the Extended Kalman Filter (EKF) algorithm. This network is able to learn the nonlinear dynamic behaviour of unknown time-varying systems and perform multi-step-ahead prediction for control purpose. The performance of this model is evaluated in modelling a multi-variable non-linear continuous stirred tank reactor (CSTR).