Multi-solution well placement optimization using ensemble learning of surrogate models

Well location optimization aims to maximize the economic profit of oil and gas field development while respecting various constraints. The limitations currently available well placement workflows are their 1) high computational requirements, which makes them inappropriate for full-field applications where a large number wells have be optimized using computationally expensive simulation model; 2) providing single optimal solution, whereas on-site operational problems often add unforeseen constraints that result in adjustments this optimal, inflexible scenario degrading its value. This study presents multi-solution, surrogate models (SMs)-assisted framework deliver diverse, close-to-optimum scenarios at reasonable cost. Simultaneous Perturbation Stochastic Approximation (SPSA) algorithm is used as optimizer diversity solutions achieved by multiple, parallel runs with different starting points. Convolutional Neural Network (CNN) SM, partly substitute reservoir model during process. A new, adjusted Latin Hypercube Sampling (aLHS) procedure developed generate initial training datasets diverse boundaries spacing An ensemble CNNs pre-trained generated dataset enhance robustness modeling allow estimation SM's prediction quality new data adaptively updated process selected points, improve accuracy. To best our knowledge, first application learning strategy problem. added value demonstrated comparing three approaches on Brugge Egg benchmark case studies. ‘no SM’: actual only, ‘Offline performed SM-only model, 3) ‘Online model. surrogate-assisted approach substantially reduced computation time, greater objective was employing adaptive due enhanced accuracy SMs. Multiple were obtained locations but values, allows more efficient exploration search space significantly presented workflow integrates critical challenges correlated, yet addressed independently, much-required flexibility efficiency operators when selecting from scenarios. • surrogate-assisted, multi-solution optimization. (SM) process, runs. several space,