Edge AI-Enabled Autonomous Well Testing with Hybrid Virtual Flow Metering for Production Surveillance

Abstract

Abstract Production surveillance is fundamental to sustaining reservoir performance, optimizing production, and enabling informed operational decision-making across large upstream assets. Conventional well-testing programs, however, remain highly dependent on manual initiation, operator experience, continuous monitoring, and post-test validation, resulting in inconsistent data quality, low test-success rates, and inefficient use of scarce testing infrastructure such as multiphase flow meters and test separators. This paper presents the development and phased field deployment of an Edge Artificial-Intelligence (AI) enabled Autonomous Well Testing framework integrated with Hybrid Virtual Flow Metering (VFM) for production surveillance in one of the world's largest onshore oil-producing assets. The solution combines machine learning, real-time edge analytics, operational-technology (OT) integration, and automated decision support to improve well-test execution, quality assurance, and production-data reliability while progressively reducing manual intervention. The framework continuously evaluates well readiness, operating stability, and test-quality conditions using both real-time and historical production data. It supports assisted, semi-autonomous, and fully autonomous testing modes, and dynamically determines whether a test should continue, be extended, be shortened, or be aborted. Automated failure diagnostics, keyed to pre-agreed rejection criteria, provide the operator with the probable cause of failure and a recommended corrective action, improving transparency and accelerating decision-making. In parallel, a hybrid virtual flow metering model estimates well production from routinely available surface measurements and is automatically recalibrated whenever a validated physical test becomes available, targeting an accuracy within 5% of the reference multiphase meter. A proof-of-concept was implemented on a representative well-gathering platform and subsequently scaled. Field results demonstrated a 50% reduction in failed well tests, an approximately threefold increase in testing throughput, test-duration reductions to under 24 hours for stable wells, and approximately 200 operational hours saved per month through early termination of non-productive tests. AI prediction accuracy approached 85% against reference measurements, and the hybrid VFM enabled a 10–15% reduction in physical testing frequency while maintaining surveillance coverage. The results demonstrate that autonomous well testing is technically feasible, operationally beneficial, and scalable across large upstream assets. The paper details the architecture, AI methodology, stability-assessment logic, VFM calibration approach, OT-cybersecurity considerations, quantified results, and a roadmap toward closed-loop autonomous field operations.