DETECTION OF UNEXPECTED EVENTS IN OIL WELLS USING MACHINE LEARNING WITH LOCAL OUTLIER FACTOR

In the oil and gas industry, investigating anomalies is crucial to prevent production losses, avert environmental accidents, and reduce maintenance costs. This study explores a density-based unsupervised machine learning model, the Local Outlier Factor (LOF), to detect anomalies in subsea production/injection wells through sensors. Experiments were conducted on real cases of wells with operational failures, focusing on all nine well anomalies defined in the public 3W database, with its most recent update called 3W 2.0. After developing the model, the proposed approach achieved satisfactory performance, with an average F1 score of 0.9511 for the nine anomalies studied, a result superior to the ones offered by other existing models in the literature. These results confirm that the proposed model, despite being a simple architecture, is effective in identifying unexpected events in a time series.