Application of Reinforcement Learning for Well Location Optimization

Abstract The extensive deployment of sensors in oilfield operation and management has led to the collection vast amounts data, which turn enabled use machine learning models improve decision-making. One prime applications data-based decision-making is identification optimum well locations for hydrocarbon recovery. This task made difficult by relative lack high-fidelity data regarding subsurface develop precise support Each placement decision not only affects eventual recovery but also decisions affecting future wells. Hence, there exists a tradeoff between maximization information gain. Existing methodologies wells during early phases reservoir development fail take an abiding view maximizing profitability, instead focusing on short-term gains. While improvements drilling technologies have dramatically lowered costs producing from prospects resulted very efficient operations, these advancements sub-optimal haphazard can lead considerable number unprofitable being drilled which, periods low oil gas prices, be detrimental company’s solvency. goal research present methodology that builds models, integrating geostatistics flow dynamics, determine A deep reinforcement (DRL) framework been proposed address issue long-horizon DRL agent employs intelligent sampling utilizes reward based geostatistical simulations. implemented approach provides opportunities insert expert while basing collected seismic prior tests. Effects are explored developed derived policies compared single-stage optimization methods development. Under similar framework, sequential strategies using shown perform better than simultaneous several