DATA PROCESSING TECHNIQUE FOR IDENTIFYING RESERVOIR FLUID INFLOW IN MANAGED PRESSURE DRILLING
R. E. Shcherbakov1, I. V. Matveev2
1Nedra Digital, St. Petersburg, Russia
2National Research Tomsk Polytechnic University, Tomsk, Russia
Keywords: machine learning, anomaly detection, time series processing, managed pressure drilling, reservoir fluid inflow
Abstract
The proposed data processing technique is designed to detect formation fluid influx during managed pressure drilling. The approach is based on multi-stage enrichment of time series with context-dependent features, which enables anomaly detection based on the dynamics of process parameters within the context of the ongoing technological operation. The method remains robust to variations in drilling operations and conditions. Experimental results demonstrate a statistically significant 13% relative increase in balanced accuracy achieved with the developed features.
|
