Predictive warning system for nonlinear process plants

Abstract

A robust warning generation method for non-linear systems is presented for forecasting abnormal situations in process systems. Predictive signals were used to issue warning in contrast to traditionally used measured states of the process. Potential unsafe conditions were identified when the most aggressive controller is unable to keep the system within normal operating range, and one or more safety constraints are violated. Conversely, process is defined in safe mode when all the output and input constraints are obeyed by the system. A moving horizon predictor is used to predict the open-loop response of the process for a defined horizon. A global optimizer seeks a sequence of feasible inputs that can keep the closed loop response of the system within the safety limits. A process is considered to be in a normal state when a sequence of feasible inputs can be found that is able to keep the process outputs within the safety limits. The efficacy of the proposed method is demonstrated on a continuous stirred tank reactor (CSTR) with different disturbance scenarios. The results show that the proposed method is able to detect a violation of a safety limit significantly earlier compared to the traditional monitoring schemes based on alarms on the measured signals.