Statistical learning based predictive models to assess stress changes-reservoir deformation due to CO2 sequestration into saline aquifers

Abstract

Mechanical integrity of saline reservoirs is necessary to ensure the safe and effective storage of CO2. In this work, three poroelastic metrics, total stress changes (horizontal and vertical) and reservoir displacement, were considered to evaluate the mechanical performance of a sandstone aquifer with a single reservoir and overlying caprock and a carbonate aquifer comprising multiple reservoirs and intermediate caprock units in the Northern Appalachian basin. While the coupled multiphase flow - geomechanics simulations could be computationally expensive, the statistical learning based simplified predictive model is used as an alternative method to assess the poroelastic performance of saline aquifers during CO2 injection. A combination of experimental design and a response surface model (proxy model) is used to develop a simplified predictive model to evaluate the poroelastic response of carbonate and sandstone reservoir caprock systems using combinations of different predictors. In this work, Box Behnken experimental design combined with polynomial proxy models were used to evaluate mechanical performance metrics using five poroelastic predictors. The predictive models, developed in this work, provide a simpler and quicker tool to evaluate the mechanical performance of the deep aquifers with similar characteristics to the reference cases. The overall conclusion of the work is that each metric parameter has its own control parameters. In addition, carbonate reservoirs with multiple caprocks use different control parameters than the sandstone reservoir. In this work, we also developed a simple linear regression model to predict surface uplift using top vertical reservoir displacement data.