Simultaneous on-line monitoring and wave-net learning

Abstract

Current on-line wave-net learning algorithm adapts the primary identified process model with the new changes in time varying processes without a consideration of abnormal situations in the process operation. Therefore, if a disturbance occurs and makes changes in the process, current on-line learning updates the primary model to an unsuitable model. This paper proposes a procedure that first determines normal variations of time-varying processes from abnormal variations incorporating an adaptive dynamic principal component analysis (Adaptive DPCA) and updates the model only based on normal variations. A double continuously stirred tank reactors (CSTR) case study is invoked to show the effectiveness of the proposed approach. The results show the effectiveness of the method.