Mai Chen

To address the liquid-loading problem at the bottom of deep gas-reservoir wells, a gas-driven jet pump for downhole deliquification was designed, and its key geometric parameters were optimized. Using high-pressure natural gas as the motive fluid, the device achieves stable liquid lifting and continuous drainage through efficient gas–liquid mixing. In this study, the internal flow characteristics of the gas jet pump were numerically investigated using CFD simulations within a steady VOF-RSM multiphase turbulence framework. Published experimental data for a prototype from the open literature were used to validate the numerical model, and the maximum relative deviation between simulation and measurement was 17.5%, satisfying the accuracy requirements for engineering CFD analysis. The effects of three key structural parameters—area ratio, nozzle-to-throat distance, and throat length-to-diameter ratio—on the liquid production rate and pumping efficiency were systematically examined through single-factor sensitivity analysis combined with an orthogonal experimental design. Range analysis showed the following order of influence: area ratio > throat length-to-diameter ratio > nozzle-to-throat distance. The optimal parameter combination was determined as an area ratio of 0.0494, a throat length-to-diameter ratio of 5, and a nozzle-to-throat distance of 7.2 mm. Under this optimized configuration, the maximum liquid production rate reaches 80 m³/d. To achieve a target liquid production rate of 20 m³/d, the required gas injection rate is reduced by 33%, and a peak pumping efficiency of 7.2% is achieved at gas injection rates between 2000 and 3000 m³/d.

Plunger lift systems are vital for optimizing production in gas wells, but their performance can be compromised by various operational anomalies. Traditional diagnostic methods and conventional convolutional neural network (CNN) approaches often struggle with the complex, transient data from these systems, particularly in capturing long-range temporal dependencies and generalizing from limited, imbalanced datasets. This study presents an enhanced diagnostic framework for plunger lift anomaly detection by leveraging the strengths of a pre-trained Vision Transformer (ViT). The methodology transforms one-dimensional time-series pressure data into two-dimensional image representations using the element-wise summation of Gramian Angular Summation Field (GASF) and Gramian Angular Difference Field (GADF), which simultaneously preserves global operational trends and local transient dynamics for vision model analysis. The ViT model, initialized with pre-trained weights, is further optimized using Bayesian optimization (BO) for hyperparameter tuning, and a tailored data augmentation pipeline is employed to improve robustness. Comparative evaluations demonstrate that the proposed ViT-based approach, particularly the ViT + GAF + BO model, significantly outperforms baseline CNN models and their optimized variants, achieving the highest Precision, Recall, and F1-score, with an F1-score of 0.93. Visualizations using t-SNE confirm the ViT’s superior capability in learning discriminative features, showcasing well-separated clusters for different operational conditions compared to CNNs. This research underscores the potential of pre-trained ViTs combined with appropriate data representation and optimization techniques for achieving accurate and reliable anomaly detection in plunger lift systems.

With the deepening of unconventional oil and gas resource development, the gas–liquid two-phase flow phenomenon in horizontal gas wells is becoming increasingly complex. Accurate and efficient prediction of the flow state has become key to optimizing production. While traditional numerical simulation methods are highly accurate, their long calculation times make them unsuitable for real-time applications. Conversely, purely data-driven methods struggle with accuracy under sparse data conditions. This paper proposes a deep operator network method (PI-DeepONet) that integrates physical prior knowledge—specifically the drift-flux model—to rapidly predict two-phase flow parameters. By jointly training the network with both data loss and physical loss, the model’s accuracy and generalization are significantly enhanced. Comparing the results with the OLGA numerical simulator verifies the model’s high performance. The average relative error of the PI-DeepONet on the test set is less than 1%, with the error of some physical quantities controlled within 0.2%. Critically, the single prediction time is less than 0.1 s, achieving a calculation speed nearly 50,000 times higher than the traditional numerical simulation method. The model significantly improves prediction speed while ensuring accuracy, making it ideal for real-time simulation and rapid response requirements in horizontal wells. This study provides a new path for intelligent diagnosis and prediction of underground working conditions and demonstrates broad engineering application potential.