Abstract This paper presents a pioneering approach to autonomous gas-lift optimization using AI-enabled Multivariable Predictive Control (MPC), aligned with the upstream oil and gas sector's digital transformation strategy. The solution integrates data-driven and machine learning models to enable real-time adaptation and optimization of oil-producing wells, targeting enhanced production with reduced gas-lift energy consumption. Automated workflows support cross-functional collaboration and streamline operations. Dynamic linear well models were developed using step test data from selected wells. A clustering technique based on choke orifice, productivity index among other parameters (Hernandez et al., 2024 1) enabled model parameter generalization across similar wells, minimizing the need for individual step tests. Inferential models were deployed to estimate bottom-hole pressure and oil rate in the absence of direct measurements. Advance Process Control (APC) strategies were implemented on a centralized Level 3 control server, integrated with the Distributed Control System (DCS), while cloud-hosted workflows managed model updates and operating envelope adjustments. The system-maintained gas-lift at the inflection point or up to a customizable economic limit (GAIN FACTOR) in terms of BLS per MMSCF, achieving maximum energy efficiency and up to 1–2% increase in overall production. A supervisory cloud layer continuously tracked soft sensor outputs and field controller data, triggering automatic recalibration workflows when deviations occurred. Digital dashboards and decision support tools provided real-time insights and well-specific recommendations, enhancing operational agility and reducing downtime. This study validates the potential of AI-driven optimization to automate complex field operations, reduce human intervention, and enable scalable, closed-loop control across gas-lifted assets demonstrating a tangible pathway for digital transformation in upstream oil and gas operations.
Rubio et al. (Mon,) studied this question.