Bayesian Biclustering by dynamics: A clustering algorithm for SAGD time series data

Abstract

Steam-Assisted Gravity Drainage (SAGD) is an extra heavy oil recovery process consisting of an upper horizontal steam injector and a lower horizontal producer that removes fluids from the reservoir. Given its costs and environmental impact, SAGD injection strategies must be examined to find ways to improve performance. In this paper, a new Bayesian Biclustering by Dynamics (BBCD) method is proposed, which finds and differentiates groups of SAGD wells based on their oil production response to steam injected over time. This is a greedy algorithm that automatically clusters both rows and columns of SAGD injection and production data and then generates a descriptive summary for each cluster. Clusters are described with probability distributions that capture the likelihood of transitioning between discrete steam-to-oil ratio states. In addition, BBCD incorporates background knowledge on SAGD directly into the clustering process via prior transition probability distributions. Real SAGD operation data from sites in Alberta, Canada are used for this analysis. The results reveal nine production responses to two different steam injection strategies, and new insights into SAGD process performance.