An Lstm-Based Anomaly Detection on Subsea Oil-Producing Well

Abstrak

The oil and gas industry faces substantial operational risks from anomalous events, necessitating effective Abnormal Event Management (AEM) to mitigate production losses and safety hazards. This study presents a supervised anomaly classification approach using Long Short-Term Memory (LSTM) networks on the 3W Dataset—comprising over 2,000 real, simulated, and expert-drawn events from offshore wells. Focusing on real instances with sufficient normal-state duration, the dataset was refined and segmented using observation windows of 60, 120, and 180 seconds. The models were trained on four selected pressure and temperature features and evaluated using precision, recall, and F1-score. Comparative analysis with Recurrent Neural Network (RNN) and Gated Recurrent Unit (GRU) models shows that the LSTM model consistently performs best, achieving a peak F1-score of 92% at a 120-second window. Furthermore, event-level performance analysis highlights the LSTM model’s strengths and limitations across different anomaly types. Compared to existing supervised and unsupervised methods on the 3W Dataset, the LSTM-based approach demonstrates competitive accuracy and robustness for real-time anomaly detection in offshore oil production systems.

Kata kunci

• anomaly detection
• time series classification
• LSTM
• offshore wells
• 3W Dataset