Insights to surfactant huff-puff design in carbonate reservoirs based on machine learning modeling

Surfactants could react with adsorbates on carbonate rock surface to alter wettability from oil-wetness water-wetness, which is effective enhance oil recovery. Surfactant huff-puff treatment mostly applied for this purpose and the resulting surfactant performance outcome of complex interfacial processes. Currently, effect important parameters not completely reported contribution each parameter hard be quantified. Traditional methods optimize are time-consuming show strong dependency extensive experiments. In paper, we address these problems machine learning (ML) perspectives. Several ML models established predict Random Forest (RF) model presents better accuracy. Based RF model, apply Shapley additive explanations (SHAP) approach interpret modeling results obtain new insights provide solutions unsolved problems. Results that when tension lower than a critical value or API gravity higher value, shows obvious improvements. general, porosity, permeability, concentration positively associated performance. The trend becomes less exceeds certain level. addition, SHAP effectively decompose into individual parameter. This analysis valuable indicate optimized. It shown in proportion samples kept low level complete alteration achieved. To concentration, an innovative procedure integrating prediction Powell’s method proposed. avoid deficient superabundant concentration. With optimization, average increased 0.37 wt% 0.93 wt%. probability high-oil-recovery class improved 0.38 0.50. incremental recovery low-oil-recovery class. Our work promotes understanding by surfactants provides fast framework predict, analyze, greatly save experiment time cost.