To address the challenges of large-scale production data, complex temporal dynamics, and the difficulty in extracting key reservoir performance indicators, this study proposes an intelligent time-series analytics approach, validated using an offshore oilfield case. The methodology integrates a cascaded outlier detection framework combining the 3-Sigma rule and the One-Class Support Vector Machine (OC-SVM). The 3-Sigma rule is first used for rapid statistical screening of extreme outliers, followed by OC-SVM for nonlinear anomaly detection, enhancing the accuracy of dynamic production data preprocessing. Key indicators—including initial production capacity, decline rate, water-cut trend, and recoverable reserves—are automatically extracted through hybrid modeling combining production decline analysis and waterflood characteristic curves. Algorithm reliability is rigorously evaluated using error metrics (SSE: Sum of Squared Errors, MSE: Mean Squared Error, MAE: Mean Absolute Error, RMSE: Root Mean Squared Error) and goodness-of-fit (R²). Experimental results demonstrate that the proposed method outperforms manual extraction, achieving <10% error in daily oil production and waterflood performance curve fitting, while significantly enhancing accuracy and automation. This framework provides a robust data−driven foundation for intelligent reservoir management.

In the context of energy transition and carbon emission reduction, Carbon Capture, Utilization, and Storage (CCUS) has emerged as a critical technological pathway. CO2 flooding in low-permeability reservoirs offers the dual benefits of enhanced oil recovery and carbon sequestration. However, its production dynamics are inherently non-stationary and strongly coupled across wells due to phase transitions and gas breakthrough phenomena. Existing approaches, including compositional simulations and data-driven models, either suffer from prohibitive computational costs or fail to ensure physically consistent predictions under complex multiphase conditions. To address this issue, an end-to-end learning and optimization framework is proposed by integrating a Physics-Inspired Graph Network (PIGN) with the multi-objective optimization algorithm NSGA-II. The framework is trained on datasets generated from numerical simulations under multiple geological realizations, with well properties, inter-well connectivity, and production schemes as inputs. Spatial–temporal dynamics are captured through a hybrid architecture combining TAGConv with GRU/LSTM modules. A multi-head independent output structure is designed to separately predict gas, oil, water, and CO2 production, while well embeddings are introduced to characterize inter-well heterogeneity. Furthermore, physics-inspired constraint losses based on local trend variations, derivative responses, and smoothness are incorporated to achieve a hybrid modeling paradigm that integrates data-driven learning with physical consistency. The surrogate model is further embedded into the NSGA-II framework, where CO2 sequestration and cumulative oil production are treated as competing objectives, enabling integrated optimization of CO2 flooding and storage. Results demonstrate that, compared with GNN-GRU, LSTM, GRU, and TCN, the proposed PIGN reduces the overall MAPE by approximately 13.0%, 23.0%, 24.3%, and 31.6%, respectively. Specifically, PIGN achieves error reductions of approximately 0.8%–1.5% in gas prediction, 1.0%–10.7% in water prediction, and 19.8%–44.4% in CO2 prediction compared to baseline models. Multi-objective optimization results indicate that the knee-point solution achieves approximately a 7.5% increase in cumulative oil production and a 4.9% increase in CO2 sequestration compared to the baseline strategy. These results validate the effectiveness of the proposed method in modeling complex multiphase dynamics and optimizing production strategies.

Precise forecasting of the initial productivity rates of infill wells is essential for the effective exploitation of offshore reservoirs characterized by high water-cut. However, conventional reservoir simulation and basic machine learning models often suffer from high computational complexity and low interpretability. This research introduces a hybrid data-driven framework that combines ensemble feature engineering with a random forest model optimized through the crayfish optimization algorithm. The primary controlling factors were identified through a majority voting mechanism involving five feature selection algorithms. Subsequently, the COA was utilized to optimize the parameters of the random forest algorithm to improve its predictive robustness. The proposed EFE-COA-RF model achieves a testing MAE of 6.831 and an R² of 0.954, outperforming standard machine learning models and other optimization-based variants. The complete training process requires approximately 10.8 min, whereas the prediction time for the testing set is approximately 0.03 s. These results demonstrate that the proposed framework provides an accurate, interpretable, and efficient tool for rapid productivity evaluation in mature offshore oilfields.

Efficient optimization of carbon dioxide (CO₂) flooding and geological storage is constrained by the high computational cost of compositional simulation and the limited capability of conventional surrogates to represent inter-well coupling and phase-dependent dynamics. This study develops a connectivity-guided phase-aware graph-temporal surrogate that explicitly embeds reservoir connectivity into Topology Adaptive Graph Convolution (TAGConv), couples graph message passing with recurrent temporal modeling, and introduces phase-specific prediction branches with reservoir-behavior-inspired temporal regularization. Unlike conventional surrogate models that mainly learn temporal or input–output mappings, the proposed framework jointly represents inter-well spatial dependency, breakthrough-sensitive multiphase dynamics, and optimization-oriented reservoir responses. Multi-seed comparisons show that the proposed model achieves the lowest active-region prediction errors for Gas, Oil, and CO₂ among the evaluated recurrent, convolutional, transformer-based, and graph-recurrent baselines, with the most evident advantage for CO₂ breakthrough dynamics. When embedded into Non-dominated Sorting Genetic Algorithm II (NSGA-II), the surrogate-assisted workflow increases cumulative oil production by approximately 7.5% and CO₂ sequestration by 4.9% relative to the baseline strategy, while providing an estimated 10.1× computational speedup over direct ECLIPSE-based optimization. High-fidelity ECLIPSE re-simulation further confirms the main response trends of the selected strategy. The framework therefore provides a connectivity-aware and phase-sensitive surrogate approach for efficient multi-objective CO₂ flooding and storage optimization.