Use of low-fidelity models with machine-learning error correction for well placement optimization

Well placement optimization is commonly performed using population-based global stochastic search algorithms. These optimizations are computationally expensive due to the large number of multiphase flow simulations that must be conducted. In this work, we present an framework in which these with low-fidelity (LF) models. LF models constructed from underlying high-fidelity (HF) geomodel a transmissibility upscaling procedure. Tree-based machine-learning methods, specifically random forest and light gradient boosting machine, applied estimate error objective function value (in case net value, NPV) associated offline (preprocessing) step, preliminary models, clustering procedure select representative set 100–150 well configurations use for training. HF simulation then configurations, tree-based trained appropriate features. online (runtime) correction, Differential evolution used all optimizations. Results presented two example cases involving vertical wells 3D bimodal channelized geomodels. We compare performance our first case, 25 runs both approaches. Our method provides overall speedup factor 46 relative best-case NPV within 1% result. second fewer conducted (consistent actual practice), approach about 8. exceeds result by 3.8%.