Massive carbon dioxide (CO₂) emissions drive climate change. Injecting CO₂ into unconventional reservoirs achieves both enhanced oil recovery (EOR) and geological sequestration. However, simultaneously optimizing oil exchange ratio, CO₂ storage, and net present value remains challenging. This study develops an integrated machine learning (ML)-based framework for multi-objective optimization of CO₂-EOR. A high-resolution reservoir simulation was constructed from field data, and Latin hypercube sampling generated diverse scenarios for proxy training. Mantel's test quantified correlations between input parameters and performance metrics, showing that injection strategy strongly controls net present value, whereas geological properties dominate CO₂ storage. Three ML models—random forest (RF), support vector regression, and artificial neural networks—were evaluated, with RF selected for its superior performance on small datasets. RF was embedded into an improved non-dominated sorting genetic algorithm Ⅱ, enhanced with grey difference degree, crowding distance, and adaptive differential evolution to improve diversity and efficiency. Finally, the technique for order preference by similarity to ideal solution ranked Pareto-optimal solutions through integrating oil productivity, storage, and economics. The proposed framework operationalizes simultaneous high-efficiency tight oil recovery and field-scale CO₂ geological storage, delivering quantitative design rules that embed low-carbon practice into upstream operations and advance the energy sector's greener and sustainable transition.

The precise determination of minimum miscible pressure is of great importance for CO₂ enhanced oil recovery and storage as it directly influences the efficiency of pore-scale oil displacement and CO₂ trapping. In this study, an interpretable machine learning framework is developed, enabling the reliable evaluation of nano-confined minimum miscible pressure. Four machine learning algorithms (Random Forest, Multi-layer Perceptron, Support Vector Regression, and eXtreme Gradient Boosting) are employed to accurately predict the nano-confined minimum miscible pressure of a CO₂-oil system. The results demonstrate that, excluding support vector regression, the determination coefficients for all models surpass 94%, signifying the robust predictive performance of our model. Subsequently, Shapley Additive exPlanations is used to analyze the feature importance ranking and the impact of each input feature on minimum miscible pressure in these models. Based on the interpretation results, our multi-layer perceptron model is superior in mining the input-output relationship and reflecting the petrophysical laws, rendering it highly suitable for predicting the minimum miscible pressure while considering nano-confinement. In addition, it is found that pore size significantly influences minimum miscible pressure prediction and that minimum miscible pressure decreases with decreasing pore size when the pore size is ≤75 nm. Single-factor sensitivity analysis is applied to validate the trend patterns between input features and minimum miscible pressure in the multi-layer perceptron model.