Fast Well Control Optimization with Two-Stage Proxy Modeling

Waterflooding is one of the methods used for increased hydrocarbon production. optimization can be computationally prohibitive if reservoir model or problem complex. Hence, proxy modeling yield a faster solution than numerical simulation. This fast provides insights to better formulate field development plans. Due technological advancements, machine learning increasingly contributes designing and building models. Thus, in this work, we have proposed application two-stage modeling, namely global local components, generate useful insights. We established models coupled them with algorithms produce new database. In paper, technique multilayer perceptron. The comprise Genetic Algorithm Particle Swarm Optimization. then implemented newly generated database build solutions that are close “ground truth”. results obtained demonstrate conducting acceptable accuracy. For optimized rate profiles, R2 metric overall exceeds 0.96. range Absolute Percentage Error generally reduces 0–3% as compared which has 0–5% error range. achieved reduction computational time by six times only using simulator.