Green Enhanced Oil Recovery in Sandstone Reservoirs: Experimental and Machine Learning Prediction of Imbibition Performance Using Bionanomaterial–Surfactant Nanohybrids

With the increased energy demand across the globe and the depletion of traditional oil reserves, there is a need to maximize the extraction of the oil in the mature oil reservoir, hence making it a priority in the petroleum industry. Enhanced Oil Recovery (EOR) methods are necessary to maximize the life of the field and maximize production, but economic and environmental factors tend to restrain their general usage. Surfactant flooding is a well-established enhanced oil recovery (EOR) technique, but high surfactant cost and adsorption onto reservoir rocks reduce efficiency, which increases its cost. Nanoparticles can mitigate this issue; however, conventional inorganic types are often toxic, costly, and difficult to scale. This study evaluates bio-silica nanoparticles (BSNPs) extracted from rice husks as a low-cost and eco-friendly alternative for surfactant flooding. BSNPs were produced under mild conditions, then functionalized with sodium dodecyl sulfate (SDS), and then dispersed in low salinity water (LSW) to form bio-silica grafted sodium dodecyl sulfate (BSDS) nanofluids. EOR performance was assessed through interfacial tension (IFT), emulsification, contact angle, and spontaneous imbibition tests on sandstone cores. The BSDS nanofluid achieved the greatest IFT reduction (~ 85 %) and altered the wettability of the core from oil-wet to water-wet (Contact angles reduced from 104.3 ° to 21.4 o) using a water drop. Spontaneous imbibition recoveries were 12.3 % for formation water (FW), 16.1 % for low salinity water (LSW), 21.9 % for BSNP, 28.7 % for SDS, and 38.1 % for BSDS, demonstrating a synergistic effect between BSNPs and SDS. These findings indicate that BSNP-based nanofluids can enhance oil recovery efficiently while offering an eco-friendly route to valorize agricultural waste. The findings were modeled in five machine learning algorithms that included Ridge regression, XGBoost regressor, K-Nearest Neighbor, Random Forest regressor, and support vector regression using Python, which demonstrated their ability to predict oil recovery from spontaneous imbibition experiments with R² values above 0.92. However, XG Boost outperformed others, with R² > 0.92, showing its capability in capturing the complex nonlinearities of spontaneous imbibition recovery.