Parallel Multiobjective Optimization for the Coupled Compositional /Geomechanical Modeling of Pulse Testing

This paper presents numerical simulation results of pulse testing experiments carried out at a test site of a carbon capture and geological storage project in Mississippi, USA. The primary objective of this study is to validate the effectiveness of pulse testing as a monitoring tool for detecting potential CO₂ leakage pathways with application to the test site. Detrending followed by Fourier transform is adopted to decompose sinusoidal pressure anomalies induced by a periodic injection of CO₂ into frequencies used as target parameters of history matching. The secondary objective is to calibrate the geologic model of the test site by reducing the discrepancy between observed and simulated Fourier parameters and assess uncertainties associated with the compositional brine-CO₂ flow. An assisted history matching tool that mounts global- and multi-objective evolutionary algorithms is developed, integrated with an in-house flow-geomechanics simulator, and employed to manage pulse testing simulations with a low computational cost in high-performance parallel computing environments. Grid cells in the test site are locally refined using enhanced-velocity that allows nonmatching grids on interfaces between subdomains. Experiments performed with one pulser well and two monitoring wells under steady-state conditions are considered baselines for subsequent experiments that convert one monitoring well into a production well as an artificial CO₂ leakage pathway. The difference between the pressure anomalies obtained from the baseline and leak experiments are captured as a signal of CO₂ leakage detection with reliability in the simulation results. A trade-off relationship between the matching qualities at the two monitoring wells is revealed more clearly by invoking multi-objective history matching than conventional global-objective history matching. This comparative study to investigate the significance of multi-objective optimization in subsurface characterization represents that diversity-preservation in the ensemble composed of qualified geologic models has the advantage of reducing bias for uncertainty quantification.