Automated Anomaly Detection Applied to DTS Data: A Case Study from Quest CCS

The applicability of an automated machine learning (ML) detection workflow applied on Distributed Temperature Sensing (DTS) as a technology to monitor CO2 containment, is presented in this paper. Quest, an onshore CCS commercial facility site in Alberta, Canada, that has received about 8.8 million metric tons of CO2 is the setting for this study. Temperature measurements are acquired in the three CO2 injection wells since 2015, across all depths, with a fiber optic system deployed behind casing. Over two hundred million temperature measurements were recorded during this time.

Our analysis includes the training and evaluation of the ML model to predict the temperature response and detect anomalous trends in the data. The workflow’s sensitivity is assessed using a pseudo-empirical model, where field data are combined with synthetic generated anomalies. Deviations as small as 0.5 degrees, aggregate over target depths, are expected to be detectable. Additionally, we discuss the implementation of the workflow on real time data and the insights gained from this process.