Steam Turbine Anomaly Detection: An unsupervised learning using Enhanced LSTM Variational Autoencoder

Abstract As a core equipment of thermal power generation, steam turbines incur significant expenses and adverse effects on operation when facing interruptions like downtime, maintenance, and damage. Accurate anomaly detection is the prerequisite for ensuring the safe and stable operation. However, challenges in steam turbine anomaly detection, including inherent anomalies, the absence of temporal information analysis, and the complexity of high-dimensional data, leading to limitations in existing unsupervised methods. To address these issues, we propose an Enhanced Long Short-Term Memory Variational Autoencoder using Deep Advanced Features and Gaussian Mixture Model (ELSTMVAE-DAF_GMM) for precise unsupervised anomaly detection. Specifically, by integrating LSTM with VAE, the LSTMVAE model was introduced to project the high-dimensional time series data to a low-dimensional phase space. Furthermore, novel deep advanced features combine deep features and reconstruction errors from LSTMVAE, significantly enhancing anomaly detection performance by representing comprehensive information and exhibiting nonoverlapping distributions in phase space. Additionally, a sample selection mechanism is developed to eliminate anomalous samples during training, improving the model precision and reliability. The study utilizes real operating data from steam turbines and conducts both comparison and ablation experiments, demonstrating superior anomaly detection outcomes characterized by high accuracy and minimal false alarm rates compared to existing approaches.