In field applications, the limited availability of pure CO 2 for CO 2 -based enhanced oil recovery (EOR) and carbon capture, utilization, and storage (CCUS) often necessitates injecting gas streams containing impurities. These impurities can significantly alter the CO 2 -oil phase behavior, mass transfer, and displacement front dynamics. Using Block X as a case study, this study integrated PVT experiments, slim-tube tests, chromatographic analysis, and numerical simulations to investigate the mechanisms by which impurity gases influence on the thermodynamic properties of CO 2 -crude oil systems and the evolutionary behavior of flooding fronts. The results show that: (1) N 2 exhibits the most adverse impact: a 10 mol% addition increases MMP by 2.88 MPa, suppresses extraction capacity by 2.1%, accelerates front propagation, and amplifies interfacial tension gradients (0.9–1.3 mN/m 2 ), inducing Marangoni-driven instability. (2) CH 4 exhibits similar but less pronounced effects. Even under miscible conditions, N 2 - and CH 4 -containing systems underperform pure CO 2 in front stability and recovery efficiency. (3) C 2 enhances miscibility, lowering MMP by 3 MPa per 10 mol%, stabilizing the front through a reduced Péclet number, and improving sweep efficiency by approximately 4%. (4) Field-scale simulations confirm that N 2 /CH 4 reduce displacement efficiency by 10%–13%, while maintaining CO 2 purity above 70% is recommended for stable operations. This work bridges impurity-induced thermodynamic alterations in CO 2 -crude oil systems with front-scale dynamic responses, clarifying how different gas impurities distinctly influence CO 2 flooding performance.
Lost circulation is a frequent and severe problem in drilling operations. To mitigate fluid loss, lost circulation material (LCM) formulations are widely used. However, designing these formulations in the field still relies heavily on engineering experience and trial-and-error, which makes it difficult to rapidly generate suitable formulations for specific lost-circulation conditions. To address this challenge, this study proposes an intelligent LCM formulation generation framework based on attribute embedding. This framework models the formulation design process as a sequential generation task, operating under given loss conditions. It comprises three main components: a loss-condition encoder, a material attribute encoder, and a formulation generation decoder. This structure enables an integrated representation of both loss conditions and material characteristics, facilitating sequential material selection and concentration prediction. The framework was evaluated using 70 LCMs and 407 successful field cases. The model achieved R2 values of 0.91 and 0.87 for total formulation concentration on the training and testing sets, respectively. The corresponding rank-1 material recall values were 0.97 and 0.86. Case studies further demonstrated that the generated formulations were consistent with those applied in the field. Moreover, the learned material representations supported reasonable substitutions among materials exhibiting similar physical and functional characteristics. To enhance practical applicability, supporting software was developed and subsequently tested in a field lost-circulation case. This successful test demonstrated the method's real-world utility. These results collectively indicate that the proposed framework offers an effective data-driven approach for targeted LCM formulation generation, thereby supporting formulation screening and engineering decision-making.
Tubing leakage in gas wells poses a critical risk to wellbore integrity and production safety. Localizing these leaks using wellhead acoustic signals, without requiring a shutdown, offers a promising solution. This study proposes an automatic tubing leakage localization method based on multi-scale physically constrained wavelet transform (PCWT) and an autocorrelation localization technique. This approach transforms the conventional manual extraction of characteristic peaks into an automatic, physics-guided decision process for characteristic delays. First, a low-frequency branch provides anchor constraints, and a composite level score (CR), which considers multiple influencing factors, is introduced to evaluate the decomposition levels of the wavelet transform. Subsequently, candidate level fusion and a cross-scale negative peak density prompting mechanism are constructed to suppress spurious time-delay characteristic peaks and valleys. The method was evaluated using 50 acoustic records collected at 5 leakage positions on a full-size wellbore leakage experimental device. It was compared against four representative methods and 4 ablation variants. PCWT achieved a failure rate of 4.0%, with position-level average relative errors of 3.67% for leakage-location estimation and 0.63% for annulus-length estimation. These results demonstrate that the proposed method can facilitate the transition of tubing leakage localization from manual interpretation to automatic identification, offering a novel technical approach for leakage monitoring and integrity management in complex wellbore conditions.
Formation Micro-Imager (FMI) data are critical for hydrocarbon exploration. However, variations in borehole size frequently cause gaps in FMI images, which hinder well-log interpretation. This study proposes InpAS, a conditional generative deep-learning model for image inpainting that incorporates stratigraphic constraints. InpAS integrates a customized Segment Anything Model (F-SAM) for gap detection with a diffusion network featuring a Horizontal-Cross-Vertical-Relating (HC-VR) attention module guided by Walther's Law. Ten-fold quantitative evaluation using Fréchet Inception Distance (FID), Inception Score (IS), Structural Similarity Index Measure (SSIM), RP, peak signal-to-noise ratio (PSNR), and mean absolute error (MAE) demonstrates that InpAS outperforms baseline models in reconstructing stratified FMI images. Furthermore, the reconstructed images exhibit superior retention of fine-grained visual details, textures, and geological continuity. These results demonstrates that integrating geological knowledge with geophysical logging and deep-learning algorithms significantly enhances the reliability of FMI gap-filling.
The rational design of hydraulic fracture parameters is crucial for the efficient development of shale gas development. However, conventional optimization approaches are often computationally expensive, and unconstrained data-driven methods may yield physically impractical designs. This study introduces a physics–data synergistic optimization framework that integrates engineering constraints, a Tent-SSA-BPNN production surrogate model, and net present value (NPV) maximization, using a dataset of 280 samples comprising 105 field-derived samples from 35 Longmaxi Formation wells and 175 samples generated by calibrated numerical models. The principal findings are as follows: (1) Five-fold cross-validation yields an average R2 of 0.91 and an RMSE of 3.1514 × 107 m3, demonstrating the predictive capability of the surrogate model; (2) For a representative 800 m horizontal lateral, the optimal design consists of 19 fracturing stages, a fracture conductivity of 90 mD·m, and a fracture half-length of 100 m; and (3) Compared with the representative field design, the optimized scheme increases the NPV by 13.8% to approximately 1.32 × 108 CNY while reducing the number of fracturing stages from 25 to 19. These results demonstrate that integrating physical feasibility, rapid production prediction, and economic evaluation provides an effective and practical approach for optimizing hydraulic–fracture parameters in shale gas horizontal wells.
