Coupling Genetic algorithm and Random Forest for robust prediction of CO2 storage efficiency in underground formations

Carbon capture and storage (CCS) is crucial to reaching net-zero emissions globally. However, uncertainty in geological CO2 storage ability forecasts is a key impediment to CCS. The majority of studies predict CO2 trapping using reservoir modeling. This method requires many computer resources to analyze a lot of subsurface data, which is costly for CO2 storage performance. This paper builds a robust machine-learning model to forecast CO2 trapping in saline aquifers with high precision to overcome a reservoir modeling restriction. This method uses a genetic algorithm (GA) and random forest concepts (RF). We acquired 1911 simulated data samples from the literature to ensure our technique was efficient and viable. These data samples were utilized to train and assess the intelligent models we provided (GA-RF). The results reveal that the proposed models' CO2 trapping performance is outstanding and acceptable. The GA-RF outperforms machine learning (ML) approaches in statistical prediction performance for measuring CO2 trapping efficiency in reservoir saline aquifers. The ML model performed well in reservoir simulations using Sleipner benchmark datasets. Our model was able to match reservoir simulation results with GA-RF predictions. The suggested robust machine learning system can assess CO2 storage operations' feasibility.