An Integrated Machining Learning-Based Workflow for CO2 Sequestration Optimization under Geological Uncertainty

Abstract

: Carbon dioxide capture and sequestration has attracted widespread interest worldwide due to greenhouse effect. Geological uncertainties affect final decisions of the injection work. Optimizing injection work under geological parameters can maximize the carbon dioxide injection efficiency and minimize the difference between the carbon dioxide storage target and actual injection volume. This work introduces an optimization workflow for decisions. It is composed of three steps. At first, generating samples as the initial data sets by using Latin Hypercube Sampling method. Secondly, a data-driven model is deployed to simulate the fluid movement in the reservoir using the samples generated in step 1. The surrogate model is optimized by tuning hyper parameters in neural networks and applying K-fold validation, which can mitigate the limitations of high-fidelity simulations. After optimization, the surrogate model is validated using full reservoir simulation. At last, with the help of genetic algorithm, both the critical pressure area and CO2 plume area reduce largely, and CO2 injection volume increases by 115*103 m3. This optimization can largely enhance CO2 sequestration efficiency. It introduces an efficient workflow to provide a reference to the decision-making process of CO2 injection location.