Alarm Monitoring for Condensate System in Thermal Power Plants Based on Bayesian Estimation

In a thermal power plant, there are numerous pieces of equipment and hundreds of process variables involved, making it difficult for the monitoring system to track all variables simultaneously. Operators determine the key variables that best reflect the condition of the system or equipment based on their experiential knowledge, then analyze data from normal conditions and build a model to obtain estimated values of variables, and the process may be evaluated by the estimation error. This paper employs an AutoRegressive Moving Average model with exogenous variables(ARMAX) to model historical data, yielding more accurate estimated values of variables and designing alarm thresholds. The Bayesian method is used to assess the reliability of estimated alarm thresholds. Used existing Bayesian methods assume a continuous uniform prior distribution, this method introduces a beta distribution to consider more information of prior knowledge. When a new observed data available, the parameters of the Bayesian method can be updated to adapt to the slow change of the operating environment. Practical industrial cases demonstrate the reliability of this method.