Anomaly Detection of Boiler Feed Pump Using Long Short Term Memory Based Auto Encoder in Power Plant Industries

In power plants, boiler feed pump plays a crucial role. As part of a prognostic and health management strategy, machine learning frameworks are currently being developed for anomaly detection to improve efficiency and optimization. The goal of this study is to create a predictive maintenance program management framework that makes use of an anomaly detection system and a data-driven model for anomaly detection that uses real-time data from the DCS system. This research makes use of an anomaly detection framework built on an LSTM-AE, or long short-term memory autoencoder neural network. The LSTM component addresses the gradient reduction issue, while the autoencoder enhances the model’s generalization capabilities. The proposed model successfully detects anomalies prior to the occurrence of event sequences.