Wen Deng

Oscillatory boundary-pressure forcing is used in pore-scale experiments and pore-network models to represent vibration-disturbed multiphase flow, but the conditions under which it reproduces distributed inertial forcing remain uncertain. This study develops a quantitative framework for assessing this surrogate in hydrologic pore networks. A throat-centered dynamic pore-network model links local residual-phase mobilization and interfacial events to network-scale residual saturation and effective permeability. The framework examines a limiting pressure-potential equivalence condition and compares throat-scale, axial-gradient, cycle-averaged power, and response-informed local root-mean-square matching. Paired calculations assess errors in both hydraulic responses, while complementary analyses examine numerical verification, pore-scale model-CFD agreement, event-closure uncertainty, finite pore-body storage, dimensionless forcing parameters, and variability across network realizations. The purpose is to identify model-conditioned, tolerance-dependent screening criteria rather than a universal ranking of the two forcing representations. The framework supports more explicit interpretation of boundary-driven experiments and simulations intended to represent vibration-induced residual-phase remobilization and hydraulic recovery.

Vibration can alter the displacement of confined non-wetting interfaces in porous media, but the resulting trajectories are difficult to interpret because slowly varying migration and vibration-induced fluctuations are coupled within a single response. Here, a confined non-wetting ganglion in an asymmetric pore throat is considered as a controlled reduced-order benchmark for investigating mixed temporal responses at the pore-throat scale. A physics-guided target transformation (TT) framework is developed, where the original displacement trajectory is represented by advancement and oscillatory components, predicted separately, and additively reconstructed under a momentum-balance constraint. The TT framework is compared with direct raw-target physics-informed prediction and an independent Hankel dynamic mode decomposition (Hankel-DMD) analysis. The results show that the TT framework achieves lower held-out trajectory prediction errors than raw-target prediction under the investigated conditions. Hankel-DMD independently recovers similar slowly varying and oscillatory temporal features from the displacement trajectories, indicating consistency between the proposed representation and an independent modal decomposition. Feature attribution analysis further identifies different input dependencies of the two component predictors within the investigated parameter domain. The proposed framework provides a physics-guided strategy for representing and predicting vibration-driven confined interface responses within a pore-throat-scale reduced-order benchmark.

Accurate production prediction of fractured horizontal wells is important for productivity evaluation and development-plan adjustment in tight oil reservoirs. However, newly producing wells often have short production histories, while production behavior varies significantly among wells due to reservoir heterogeneity, stimulation differences, and production operations. This study proposes a similarity-guided sequential transfer learning framework for short-history production prediction. Candidate source wells are selected by jointly considering inter-well distance, static geological-engineering similarity, and dynamic production similarity. Their production dynamics are then sequentially transferred into a target-well gated recurrent unit (GRU) model along optimized transfer paths, rather than directly mixing multi-well samples. The framework was evaluated using production data from 40 hydraulically fractured horizontal wells in a tight oil reservoir block of the Daqing Oilfield and compared with the Arps decline model, tree-based models, back-propagation neural network (BPNN), and conventional GRU. Results show that the proposed method improves prediction stability under limited target-well histories, with 30 target wells achieving high-quality predictions and only two classified as low-quality predictions. Abnormal-well diagnosis indicates that persistent prediction bias mainly results from low source-target transferability. Data-efficiency analysis shows that approximately 20 months of production history provides initial engineering usability, while 25 months or longer leads to more stable prediction. The proposed method provides a practical tool for new-well production forecasting and dynamic adjustment of tight oil development plans.

Trapped non-wetting droplets can fundamentally alter pressure transmission in vibrated porous media. Unlike the relatively regular response of single-phase systems, oscillating trapped droplets act as dynamic boundaries that induce complex local pressure redistribution through their coupling with bypass flow. To explicitly characterize this mechanism, we develop a pore-doublet based dynamic boundary-pressure expression for trapped droplets under vibration. The model consists of a constricted trapping channel and a parallel bypass channel, representing the essential hydraulic interaction between droplet motion and pressure redistribution. Focusing on the pre-mobilization stage, we derive an explicit time-dependent pressure expression and assess its cross-model consistency through comparison with two-phase CFD simulations. Good agreement is obtained for droplet trajectories and critical mobilization thresholds over the low-to-intermediate frequency range. The expression is further embedded into a pore-network model as an initial demonstration of upscaled application. The proposed formulation provides a reduced-order and physically interpretable bridge between pore-scale trapped droplet induced pressure redistribution and larger-scale multiphase seepage evolution.

High-fidelity simulation of vibration-driven pore-scale interfacial dynamics is expensive, whereas labeled data in new operating regimes are typically limited. To address this challenge, this study develops a controlled-accuracy data-efficient learning framework for interfacial-displacement prediction under geometry and forcing shifts. Multi-stage sequential transfer learning (STL) is used to progressively adapt a predictor from a source domain to target domains with limited labeled data. A controlled-accuracy criterion, defined by a prescribed relative-error threshold and an acceptance ratio of at least 90%, is introduced to quantify the minimum labeled-data demand through 𝑁required and Δ𝑁/𝑁. Results show that, under the same engineering accuracy requirement, STL significantly reduces target-domain labeled-data demand while maintaining reliable prediction performance, and CNN-LSTM-STL achieves better robustness in strongly nonlinear regimes. The minimum controlled-accuracy labeled-data demand decreases by approximately one order of magnitude at the first transfer stage and is further reduced by about 50% in subsequent stages. Overall, the proposed framework provides an effective data-efficient learning strategy for nonlinear pore-scale dynamics under progressive domain shift.