Multi-task learning for virtual flow metering

Virtual flow metering (VFM) is a cost-effective and non-intrusive technology for inferring multiphase rates in petroleum assets. Inferences about are fundamental to decision support systems that operators extensively rely on. Data-driven VFM, where mechanistic models replaced with machine learning models, has recently gained attention due its promise of lower maintenance costs. While excellent performances small sample studies have been reported the literature, there still considerable doubt robustness data-driven VFM. In this paper, we propose new multi-task (MTL) architecture Our method differs from previous methods it enables across oil gas wells. We study by modeling 55 wells four assets compare results two single-task baseline models. findings show MTL improves over methods, without sacrificing performance. yields 25%–50% error reduction on average architectures struggling. • Large-scale virtual meter Multi-task better adhere physical expectations. Sharing data can improve performance challenging cases.