Reservoir permeability serves as a critical parameter for unconventional reservoir characterization and hydrocarbon recovery optimization. However, complex petrophysical mechanisms and multifactorial coupling make its seismic prediction face significant challenges. This review comprehensively synthesized advances and limitations across three dominant methodologies: (i) dispersion/attenuation-based methods, limited by petrophysical assumptions, scaling issues, and non-uniqueness; (ii) pore structure-constrained methods, enhancing prediction accuracy but hindered by oversimplification and high-dimensional inversion instability; and (iii) artificial intelligence frameworks, offering data efficiency yet challenged by error propagation, overfitting vulnerability, and geologically implausible extrapolation. Comparative analysis revealed core bottlenecks in inadequate multiscale coupling between petrophysical mechanisms and data-driven approaches. These challenges are compounded by the absence of cross-disciplinary validation frameworks. To address these challenges, this review integrated interdisciplinary perspectives from seismic exploration, petrophysics, and machine learning. It proposed a tripartite permeability prediction paradigm unifying physical mechanisms, data-driven techniques, and engineering validation. This framework encompasses: first, advancing multi-porosity fluid-solid coupling theory and pore structure-constrained rock physics models; second, constructing physics-guided multimodal learning architectures that deeply embed differentiable physical laws (e.g., Darcy-Biot theory) within cross-scale physics-informed neural networks, coupling microscopic pore network simulations with macroscopic seismic responses; third, establishing a closed-loop workflow covering digital rock core simulations, blind well testing validation, production history matching, and dynamic data-driven evolution, thereby forming a quantifiable and iteratively upgradable technological system. This paradigm provides a multiscale approach for accurately characterizing permeability in unconventional reservoirs, and it establishes foundational theoretical principles and delineates practical implementation pathways for economically viable unconventional resource development.

Fiber optic distributed acoustic sensing (DAS) based on phase-sensitive optical time-domain reflectometry holds significant potential for monitoring applications in seismic exploration, pipeline integrity, and border security. Conventional straight-fiber DAS systems are inherently limited to detecting single-component vibration signals along the fiber axis. To address this limitation, we propose a distributed helically wound cable (HWC). In this article, we present a theoretical analysis of the fundamental mathematical model governing HWC response and the selection criteria for an optimal spiral wrapping angle. We conducted a pioneering three-dimensional seismic field experiment in Xinghua, Jiangsu, China. An innovative underwater cable deployment scheme was implemented to ensure effective coupling between the cable and the surrounding medium. Experimental results demonstrated that HWC with a 30° wrapping angle yielded single-shot records characterized by a high signal-to-noise ratio and a broad effective frequency bandwidth, and enabled clear identification of shallow reflection events in stacked sections. This confirms the capability of HWC to acquire ground seismic reflection signals. Our findings provide an effective pathway for advancing next-generation fiber optic distributed seismic exploration technology.

Understanding the relationship between micro-cracks and elastic anisotropy is crucial for characterizing subsurface flow pathways, optimizing hydraulic fracturing, and enhancing seismic interpretation in unconventional shale reservoirs. Although clay content and total organic carbon (TOC) are recognized primary controls on anisotropy, the specific influence of sedimentary structures on micro-crack parameters (such as crack porosity, crack density, and aspect ratio) and their contribution to anisotropic behavior have not been fully quantified, particularly in lacustrine shales with varied sedimentary architectures. In this study, 17 shale samples were categorized into three sedimentary structural types: laminated, bedded, and massive, based on their microstructure characteristics. Ultrasonic velocity measurements were performed on 17 paired shale plugs under confining pressures to quantify the relationship between micro-crack parameters and elastic anisotropy. Experimental results reveal a clear difference in stress sensitivity of bedding-normal velocities: Laminated shales > bedded shales > massive shales, which are attributed to varying degrees of micro-crack alignment and density. Laminated shales exhibit the strongest anisotropic properties, followed by bedded shales, while massive shales show weak anisotropy. Velocity predictions from the Mori-Tanaka effective medium model are in good agreement with the measurements, validating its applicability for shales with diverse structures. Micro-crack analysis indicates a positive correlation between crack density/porosity and anisotropy magnitude. Notably, laminated shales are characterized by the highest crack porosity (0.012-0.015%), high clay content (average 40%), and moderate TOC, indicating a combined effect of composition and microstructure on anisotropy. This study highlights that sedimentary structure plays a key role in controlling micro-crack development and related anisotropy in lacustrine shales, with laminated shales exhibiting the most significant combined effect, thus improving the accuracy of minimum-horizontal-stress prediction and hydraulic-fracture design.

Suppressing complex mixed noise in seismic data poses a significant challenge for conventional methods, which often cause signal damage or leave residual noise. While sparse basis learning is a promising approach for this task, traditional data-driven learning methods are often insensitive to the physical properties of seismic signals, leading to incomplete noise removal and compromised signal fidelity. To address this limitation, we propose a physics-constrained sparse basis learning method for mixed noise suppression. Our method integrates local dip attributes—estimated and iteratively refined by a plane-wave destructor filter—as a physical constraint within the dictionary learning framework. This constraint guides the learning process to achieve high-fidelity signal reconstruction while effectively suppressing multiple noise types. Tests on complex synthetic and real data demonstrate that the proposed method outperforms conventional techniques and industry-standard workflows in attenuating mixed noise, including strong anomalous amplitudes, ground roll, and random and coherent components, thereby significantly enhancing the signal-to-noise ratio and imaging quality.

The Hessian matrix, though computationally expensive, plays a critical role in ensuring inversion accuracy and mitigating cross-talk in multi-parameter inversion. The well-known wavefield reconstruction inversion (WRI) or extended space full-waveform inversion can reduce nonlinearity and mitigate cycle skipping in traditional FWI. However, most implementations omit the Hessian. In this study, the Hessian—formulated as a function of measurement and theoretical covariance matrices—is incorporated into WRI within a Bayesian inference framework. Furthermore, the connections between the data- and model-domain Hessian equations are discussed, leading to a simplified calculation method for the extended source. Based on this approach, a new definition for the theoretical covariance matrix is proposed and validated through numerical tests, demonstrating its accuracy.

Accurate prediction of reservoir porosity is fundamental for hydrocarbon resource evaluation and development planning, yet traditional methods struggle with spatial heterogeneity and complex geological structures. This study proposes a hybrid deep learning framework that integrates U-Net++ with an attention-guided graph neural network to simultaneously capture multiscale well logging data features and non-Euclidean spatial dependencies. The model incorporates dense skip connections, deep supervision, and dual-channel attention mechanisms to enhance both local feature extraction and global topological modeling. Experiments on a real-world continental sedimentary basin dataset (26 wells, ~40 km2) demonstrated that the proposed method achieved a mean squared error (MSE) of 4.62, mean absolute error of 1.24, coefficient of determination (R2) of 0.912, and structural similarity index measure of 0.831, representing a 14.9-38.7% reduction in prediction errors relative to widely used deep learning and graph-based baselines. Statistical tests (p<0.05) confirmed the significance of the improvements. The model was particularly robust in extreme porosity ranges (>16% or <8%), reducing errors by 23.1-42.6% compared to U-Net++. Ablation studies highlighted the contribution of graph structure (19.0% MSE reduction), attention mechanism (15.0%), and deep supervision (12.5%). Beyond predictive accuracy, attention-weight analysis revealed strong alignment with geologically meaningful features, such as faults and sedimentary facies boundaries, thereby enhancing interpretability. The proposed framework offers a scalable and interpretable solution for reservoir characterization, with broad potential applications in heterogeneous and faulted reservoirs.