The SPE DSEA AI 4 Energy Workshop is a one-day event hosted by the Society of Petroleum Engineers (SPE) Aberdeen Section in collaboration with the SPE London Section. It focuses on practical applications of Artificial Intelligence (AI) in the energy industry, specifically in Data Science and Engineering Analytics (DSEA). The workshop aims to demystify AI and show how it can be used effectively in day-to-day operations in the energy sector.
Anomaly detection pipeline for surface logging data
At the recent SPE A.I.4Energy Workshop, Fabio Concina, Kwantis’ R&D Manager, presented an anomaly detection pipeline designed to enhance the quality of surface logging data in drilling operations. This innovative approach combines traditional techniques with advanced AI to accurately detect low-quality intervals, significantly improving data integrity and operational efficiency.
The method involves a three-step process: first, sensor data undergoes preprocessing to handle missing values and normalize the data, with basic univariate anomaly detection methods used to filter out trivial anomalies based on domain knowledge. Second, a trained autoencoder neural network reconstructs inputs, minimizing errors for normal sequences and maximizing errors for anomalous ones. Third, a novel thresholding technique is applied to reconstruction errors to detect anomalous data points, which are then assigned an anomaly score and clustered to identify low-quality intervals.
A Novel Thresholding Approach for Real-Time Monitoring
The approach leverages thresholding for high precision, allowing for sensitivity adjustment, and integrates traditional methods to ensure high-quality training data. Its modular design ensures adaptability and robustness, significantly improving data quality and operational safety in drilling processes.
The study highlighted the effectiveness of combining traditional and AI-based methods and suggested future exploration of additional deep learning models to further enhance accuracy. The novel integration of traditional and AI techniques, along with the unique thresholding method, provides substantial benefits for real-time monitoring in drilling operations, offering valuable insights for engineers working in real-time operation centers.