Multi-Engine AI Tool for Oil Field Development Plan Optimization

Abstract Amount of mature fields has never been higher than today. Our objective was to develop a tool capable of fast generation of an optimized scenario of field development. Multivariate optimization workflow enhanced by AI algorithms, offers a wide assessment of field data. It is capable of proposing the most promising well control parameters and suggesting the most effective type of workovers and possible ways of well layout modifications. The input for the tool includes a very few and easy-to-obtain production parameters such as production and injection rates, pressure, etc. The tool is denoted as "multi-engine" because it uses several predictive models in the optimization workflow to gather various insights about the field from the production data and obtain the results. For example, material balance model in a form of Capacitance-Resistance model (CRM) is used for interwell connectivity field identification and enables the infill drilling optimization. As for engines, physics-driven (a modified CRM), data-driven (Recurrent Neural Network, or RNN) and hybrid (Physics-Informed Neural Network) models are used. The tool is designed to gain the computational benefits from both data-driven and physics driven models.