Anomaly Detection of Power Plant Equipment Using Long Short-Term Memory Based Autoencoder Neural Network

Abstract

Anomaly detection is of great significance in condition-based maintenance of power plant equipment. The conventional fixed threshold detection method is not able to perform early detection of equipment abnormalities. In this study, a general anomaly detection framework based on a long short-term memory-based autoencoder (LSTM-AE) network is proposed. A normal behavior model (NBM) is established to learn the normal behavior patterns of the operating variables of the equipment in space and time. Based on the similarity analysis between the NBM output distribution and the corresponding measurement distribution, the Mahalanobis distance (MD) is used to describe the overall residual (OR) of the model. The reasonable range is obtained using kernel density estimation (KDE) with a 99% confidence interval, and the OR is monitored to detect abnormalities in real-time. An induced draft fan is chosen as a case study. Results show that the established NBM has excellent accuracy and generalizability, with average root mean square errors of 0.026 and 0.035 for the training and test data, respectively, and average mean absolute percentage errors of 0.027%. Moreover, the abnormal operation case shows that the proposed framework can be effectively used for the early detection of abnormalities.

RNN-Autoencoder Approach for Anomaly Detection in Power Plant Predictive Maintenance Systems

“…Many studies have developed unsupervised learning methods to overcome the problems of unbalancing categorical training datasets [8,11,21,22]. Anomaly detection was previously performed using the pattern recognition method, which detects deviations in specific datasets using the "normal" behavior model (NBM).…”

“…NBM is designed to investigate the equipment operating variable's healthy behavior patterns under certain conditions and at particular times. This method involves rebuilding the 'typical' behavior of time-series parameter data and then detecting abnormalities using reconstruction errors [10][11][12]15]. The data used to form the NBM of this equipment is obtained from the historical operation of the equipment with special treatment, including eliminating data when equipment downtime occurs, eliminating data when a failure occurs according to operating records, and eliminating abnormal data based on statistical characteristics [11].…”

“…One model that is widely used to build NBM for anomaly detection applications is the autoencoder (AE) model, which is an unsupervised learning method in which the output of the model is a reconstruction of the input [11,14,15] with latentspace representation techniques [23][24][25]. AE architecture is widely used because it can overcome the difficulty of determining features manually, which requires domain knowledge about data and processes related to it [11,26] and can automatically extract representative features from raw data so that it significantly improves failure detection accuracy [22,27]. AE can also improve the model's generalizability with data reduction [15,28].…”

“…Maintenance activities are carried out to avoid failure events and improve equipment performance. The presence of anomaly detection allows practical maintenance tasks planning on equipment that can reduce high losses due to unplanned downtime and has great significance in the condition based maintenance (CBM) [10][11][12]. Online equipment anomaly detection will indicate equipment abnormality and initiate component inspection activities in the field [10].…”

Detecting Anomalous Transactions via an IoT Based Application: A Machine Learning Approach for Horse Racing Betting

“…One possible solution to this problem is time-series machine learning. LSTM, a time-series machine learning architecture, has been used to detect fraud in applications ranging from credit card transactions [ 31 ] to power plant abnormalities [ 34 ]. This study focused on the order of transactions, rather than analyzing each transaction event.…”