Search Results (768)

Nonlinear loads are widely applied, making the generation mechanism of grid harmonics increasingly intricate. However, high-precision monitoring devices suffer from high deployment costs and limited coverage. This poses a major challenge to directly acquiring harmonic voltages at some nodes. To solve this problem, this paper proposes a harmonic state estimation method based on a Dual-Stream Adaptive Fusion Generative Adversarial Network (DSAF-GAN), with an innovative design in its generator architecture. A dual-path generator is developed to extract multi-scale features through heterogeneous network branches collaboratively. The ResNet-GRU path integrates convolutional residual modules with Bidirectional Gated Recurrent Units (Bi-GRUs). It effectively captures local spatial patterns and temporal dynamic characteristics of time-series data. The multi-layer perceptron (MLP) path focuses on mining global nonlinear correlations, thereby enhancing the overall feature-expressing capability. An adaptive weight fusion module (Attention Weight Net) fuses the outputs of the two paths. It dynamically allocates contribution weights, improving the model’s flexibility and generalization performance. Experimental results show that the proposed DSAF-GAN can accurately reconstruct the harmonic voltage component content rate of missing nodes. Full article

Real-time object detection in Unmanned Aerial Vehicle (UAV) imagery presents unique challenges, primarily characterized by extreme scale variations and intense background clutter. Existing detectors often suffer from spectral homogenization in which the critical high-frequency details of minute targets are washed out by dominant background signals during feature downsampling. To address this, we propose Eagle-YOLO, a dynamic feature aggregation framework designed to master these complexities without compromising inference speed. We introduce three core innovations: (1) the Hierarchical Granularity Block (HG-Block), which employs a residual granularity injection pathway to function as a detail anchor for tiny objects while simultaneously accumulating semantics for large structures; (2) the Cross-Stage Context Modulation (CSCM) mechanism, which leverages a global context query to filter background redundancy and recalibrate features across network stages; and (3) the Scale-Adaptive Heterogeneous Convolution (SAHC) strategy, which dynamically aligns receptive fields with the inherent scale distribution of aerial data. Extensive experiments on the DUT Anti-UAV dataset demonstrate that Eagle-YOLO achieves a remarkable balance between accuracy and latency. Specifically, our lightweight Eagle-YOLO-T variant achieves 74.62% AP, surpassing the robust baseline RTMDet-T by 1.67% while maintaining a real-time inference speed of 141 FPS on an NVIDIA RTX 4090 GPU. Furthermore, on the challenging Anti-UAV dataset, our Eagle-YOLOv8-M variant reaches an impressive 94.38% ${\mathrm{AP}}_{50}^{\mathrm{val}}$, outperforming the standard YOLOv8-M by 2.83% and proving its efficacy for edge-deployed aerial surveillance applications. Full article

Understanding the behavior of light non-aqueous phase liquids (LNAPLs) in urban subsurface environments is essential to developing effective pollution control strategies, designing remediation systems, and managing waste and resources sustainably. Oil leakage from urban industrial facilities, underground pipelines, and fueling systems often leads to contamination that is challenging to characterize due to complex soil structures, limited access beneath densely built infrastructure, and dynamic groundwater conditions. In this study, we integrate a large-scale soil tank experiment with multiphase flow simulations to elucidate LNAPL distribution mechanisms under fluctuating groundwater conditions. A 2.4-m-by-2.4-m-by-0.6-m soil tank was used to visualize oil movement with high-resolution multispectral imaging, enabling a quantitative evaluation of saturation distribution over time. The results showed that a rapid rise in groundwater can trap 60–70% of the high-saturation LNAPL below the water table. In contrast, a subsequent slow rise leaves 10–20% residual saturation within pore spaces. These results suggest that vertical redistribution caused by groundwater oscillation significantly increases residual contamination, which cannot be evaluated using static groundwater assumptions. Comparisons with a commonly used NAPL simulator revealed that conventional models overestimate lateral spreading and underestimate trapped residual oil, thus highlighting the need for improved constitutive models and numerical schemes that can capture sharp saturation fronts. These results emphasize that an accurate assessment of LNAPL contamination in urban settings requires an explicit consideration of groundwater fluctuation and dynamic multiphase interactions. Insights from this study support rational monitoring network design, reduce uncertainty in remediation planning, and contribute to sustainable urban environmental management by improving risk evaluation and preventing the long-term spread of pollution. Full article

Physics–Informed Neural Networks (PINNs) have demonstrated efficacy in solving both forward and inverse problems for nonlinear partial differential equations (PDEs). However, they frequently struggle to accurately capture multiscale physical features, particularly in regions exhibiting sharp local variations such as shock waves and discontinuities, and often suffer from optimization difficulties in complex loss landscapes. To address these issues, we propose LocRes–PINN, a physics–informed neural network framework that integrates local awareness mechanisms with residual learning. This framework integrates a radial basis function (RBF) encoder to enhance the perception of local variations and embeds it within a residual backbone to facilitate stable gradient propagation. Furthermore, we incorporate a residual–based adaptive refinement strategy and an adaptive weighted loss scheme to dynamically focus training on high–error regions and balance multi–objective constraints. Numerical experiments on the Extended Korteweg–de Vries, Navier–Stokes, and Burgers equations demonstrate that LocRes–PINN reduces relative prediction errors by approximately 12% to 67% compared to standard benchmarks. The results also verify the model’s robustness in parameter identification and noise resilience. Full article

To address the challenge in multi-view learning within practical application scenarios—such as smart grid multi-source monitoring and complex environment perception—where view quality often exhibits significant dynamic time-varying characteristics due to environmental interference or sensor failures, rendering traditional static fusion methods inadequate for maintaining decision-making reliability, a general adaptive robust fusion method, termed the Consensus-Aware Residual Gating (CARG) mechanism, is proposed. This approach constructs a sample-level dynamic quality assessment framework. It computes three interpretable metrics—self-confidence, group consensus, and complementary uniqueness—for each feature view in real time, thereby accurately quantifying instantaneous data quality fluctuations. A multiplicative gating structure is employed to generate dynamic weights based on these metrics, embedding a structural inductive bias of group consensus priority. Specifically, when quality degradation triggers view conflicts, the mechanism prioritizes majority-consistent reliable signals to suppress noise; when high-value complementary information emerges, it cautiously incentivizes discriminative features to rectify group bias. This design achieves adaptive perception of quality variations and robust decision-making without relying on additional weight-prediction networks. Extensive experiments are conducted on general multi-view benchmarks. The results demonstrate that CARG surpasses mainstream algorithms in accuracy, robustness, and interpretability. It effectively shields decisions from anomalous feature interference and validates its efficacy as a universal fusion framework for dynamic environments. Full article

Background: Multiple sclerosis (MS) is a neuroinflammatory disease characterized by autoimmune-driven inflammation in the central nervous system that damages axons and destroys myelin. It is difficult to diagnose multiple sclerosis due to its complexity, and different people may react differently to different treatments. While the exact cause of multiple sclerosis (MS) and the reasons for its increasing prevalence remain unclear, it is widely believed that a combination of genetic predisposition and environmental influences plays a significant role. Methods: Finding biomarkers for complicated diseases like multiple sclerosis (MS) is made more promising by the emergence of network and system biology technologies. Currently, using tools like Network Analyst to apply network-based gene expression profiling provides a novel approach to finding potential medication targets followed by molecular docking and MD Simulations. Results: There were 1200 genes found to be differentially expressed, with CD44 showing the highest degree score of 15, followed by CDC42 and SNAP25 genes, each with a degree score of 14. To explore the regulatory kinases involved in the protein–protein interaction network, we utilized the X2K online tool. The present study examines the binding interactions and the dynamic stability of four ligands (Obeticholic acid, Chlordiazepoxide, Dextromethorphan, and Hyaluronic acid) in the Hyaluronan binding site of the human CD44 receptor using molecular docking and molecular dynamics (MD) simulations. Docking studies demonstrated a significant docking score for Obeticholic acid (−6.3 kcal/mol), underscoring its medicinal potential. MD simulations conducted over a 100 ns period corroborated these results, revealing negligible structural aberrations (RMSD 1.3 Å) and consistent residue flexibility (RMSF 0.7 Å). Comparative examinations of RMSD, RMSF, Rg, and β-factor indicated that Obeticholic acid exhibited enhanced stability and compactness, establishing it as the most promising choice. Conclusions: This integrated method underscores the significance of dynamic validations for dependable drug design aimed at CD44 receptor-mediated pathways. Future experimental techniques are anticipated to further hone these findings, which further advance our understanding of putative biomarkers in multiple sclerosis (MS). Full article

To solve the difficulty of autonomous decision-making caused by the complex driving environment and changeable weather conditions, an end-to-end autonomous decision-making method based on residual network (ResNet), convolutional block attention module (CBAM) and bidirectional long short-term memory network (BiLSTM) is proposed for intelligent vehicles. Firstly, ResNet is used to extract spatial feature information contained in driving scene images. Then, CBAM is adopted to assign weights to each network channel and dynamically focus on important spatial regions in the image. Finally, BiLSTM is constructed to process the contextual features of continuous scenes, and the autonomous decision-making of intelligent vehicles is achieved through the fusion of spatial features and temporal information. On this basis, the proposed network model is trained using a real-world driving dataset and fully tested in various scenarios. Moreover, ablation experiments are conducted to verify the contribution of each module to the overall performance. The results show that the proposed method has better accuracy and stability compared with multiple existing methods, including PilotNet, FCN-LSTM, and DBNet, and its accuracy reaches 90.16% under clear weather conditions, as well as 81.29% under nighttime and snowy weather conditions. Full article

With the rapid advancement of deepfake generation technologies, forged videos have become increasingly realistic in visual quality and temporal consistency, posing serious threats to multimedia security. Existing detection methods often struggle to effectively model temporal dynamics and capture subtle inter-frame anomalies. To address these challenges, we propose a Stochastic Differential Equation and Quantum Uncertainty Network (SDEQ-Net), a novel deepfake video anomaly detection framework that integrates continuous time stochastic modeling with quantum uncertainty mechanisms. First, a Continuous Time Neural Stochastic Differential Filtering Module (CNSDFM) is introduced to characterize the continuous evolution of latent inter-frame states using neural stochastic differential equations, enabling robust temporal filtering and uncertainty estimation. Second, a Quantum Uncertainty Aware Fusion Module (QUAFM) incorporates Hermitian-symmetric density matrix representations and von Neumann entropy to enhance feature fusion under uncertainty, leveraging the mathematical symmetry properties of quantum state representations for principled uncertainty quantification. Third, a Fractional Order Temporal Anomaly Detection Module (FOTADM) is proposed to generate fine grained temporal anomaly scores based on fractional order residuals, which are used as dynamic weights to guide attention toward anomalous frames. Extensive experiments on three benchmark datasets, including FaceForensics++, Celeb-DF, and DFDC, demonstrate the effectiveness of the proposed method. SDEQ-Net achieves AUC scores of 99.81% on FF++ (c23) and 97.91% on FF++ (c40). In cross dataset evaluations, it obtains 89.55% AUC on Celeb-DF and 86.21% AUC on DFDC, consistently outperforming existing state-of-the-art methods in both detection accuracy and generalization capability. Full article

With the large-scale integration of renewable energy and power electronic devices, power quality disturbances exhibit strong nonlinearity and complex dynamic behavior. Traditional methods are limited by insufficient feature extraction and cumbersome classification, often failing to meet practical accuracy and robustness requirements. To address this issue, this paper proposes a multi-level ensemble method for power quality disturbance identification. A time–frequency dual-branch feature extraction module was designed, combining residual networks and bidirectional temporal convolutional networks to capture both local discriminative features and long-range temporal dependencies in the time and frequency domains. A cross-attention mechanism was further employed to fuse the time–frequency features, enabling adaptive focus on the most critical information for disturbance classification. The fused features were fed into fully connected layers and a Softmax classifier for multi-class identification. Experimental results demonstrated superior accuracy, robustness, and generalization capability compared with existing methods, validating the effectiveness of the proposed model. Full article

Accurately forecasting agricultural commodity prices is a complex and persistent problem for producers, traders, and policymakers. In this study we examine how artificial intelligence can be combined with large-scale global news data to refine daily corn price forecasts. A Long Short-Term Memory (LSTM) neural network was trained on Chicago corn futures between 2021 and 2024 to capture price dynamics, while agriculture-related news features were derived from the Global Database of Events, Language, and Tone (GDELT). Rather than sentiment polarity, the analysis shows that attention-based indicators, such as article volume, rolling intensity measures, and persistence of elevated coverage, carry stronger predictive information. These features are incorporated through a Ridge regression residual correction applied to the LSTM predictions, forming a lightweight two-stage hybrid model. While absolute forecast accuracy remains comparable to the price-only baseline (RMSE ≈ 9 ¢/bu; MAE ≈ 5.8 ¢/bu; R² ≈ 0.99), the hybrid framework improves directional accuracy by approximately 2.4 percentage points, with gains concentrated during periods of moderate news intensity. Feature attribution results indicate that media attention intensity and persistence dominate sentiment-tone variables, which receive zero weight under regularization. Overall, the proposed framework offers a transparent, computationally efficient, and reproducible approach for integrating open global news data into short-term agricultural price forecasting. Full article

Rapid urbanization is fragmenting ecological spaces in megacities, threatening biodiversity and ecosystem services. Yet, it remains unclear whether, and under what conditions, urban ecological networks (ENs) can recover robustness once heavily disrupted. This study aims to (i) develop a dynamic assessment framework that couples network robustness and connectivity, and (ii) apply it to examine how ENs evolve under sustained urbanization and shifting policy regimes. Using multi-period data for Shenzhen, China (2000–2025), we simulate deliberate and random attacks on patches and corridors to derive data-driven thresholds that grade the importance of ecological elements, and integrate these with graph-based connectivity metrics to track changes in network structure and node centrality over time. Shenzhen’s EN exhibits a typical “fragmentation–reconfiguration–optimization” pathway, with a “rapid decline–deceleration–recovery” trajectory in robustness that closely aligns with the introduction of strict ecological control lines and subsequent restoration initiatives. The results show that targeted protection of residual core habitats, combined with strategic reconnection and infill greening in the urban interior, can reverse earlier losses in network robustness. The proposed robustness-informed framework provides operational guidance for prioritizing protection, restoration, and optimization of ecological space, and offers a transferable approach for adaptive EN planning in high-density tropical and subtropical megacities. Full article

Accurate prediction of lithium battery remaining useful life (RUL) is crucial for battery management systems to monitor battery health status. However, RUL prediction remains challenging due to capacity non-stationarity caused by capacity regeneration phenomena. Therefore, this study proposes a novel RUL prediction framework that combines a two-stage decomposition strategy with a parallel Informer-LSTM architecture. First, STL decomposition is employed to decompose the capacity sequence into trend, seasonal, and residual components. The VMD method further refines the residual component from STL, extracting the underlying multiscale subsignals. Subsequently, a parallel dual-channel prediction network is constructed: the Informer branch captures global long-range dependencies to prevent trend drift, while the LSTM branch models local nonlinear dynamics to reconstruct fluctuations associated with capacity regeneration. Experiments on the NASA dataset demonstrate that this framework achieves an MAE below 0.0109, an RMSE below 0.0160, and an R² above 0.9950. Additional validation on the Oxford battery dataset confirms the model’s robust generalization capability under dynamic conditions, with an MAE of 0.0017. This further demonstrates that the proposed RUL prediction framework achieves significantly enhanced prediction accuracy and stability, offering a reliable solution for battery health status detection in battery management systems. Full article

Polylactic acid (PLA) is a promising biodegradable polymer whose widespread application is hindered by inherent brittleness. Polyethylene glycol (PEG) is a common plasticizer, but the effects of intermediate molecular weights, such as 4000 g/mol, on the coupled thermal, mechanical, and rheological properties of PLA remain insufficiently understood. This study presents a comprehensive analysis of PLA plasticized with 0–20 wt% PEG 4000, employing differential scanning calorimetry (DSC), dynamic mechanical analysis (DMA), and rheology. DSC confirmed excellent miscibility and a significant glass transition temperature (T_g) depression exceeding 19 °C for the highest concentration. A complex, non-monotonic evolution of crystallinity was observed, associated with the formation of different crystalline forms (α′ and α). Critically, DMA revealed that the material’s thermo-mechanical response is dominated by its thermal history: while the plasticizing effect is masked in highly crystalline, as-cast films, it is unequivocally demonstrated in quenched amorphous samples. The core finding emerges from a targeted rheological investigation. An anomalous increase in melt viscosity and elasticity at intermediate PEG concentrations (5–15 wt%), observed at 180 °C, was systematically shown to vanish at 190 °C and in amorphous samples. This proves that the anomaly stems from residual crystalline domains (α′ precursors) persisting near the melting point, not from a transient molecular network. These results establish that PEG 4000 is a highly effective PLA plasticizer whose impact is profoundly mediated by processing-induced crystallinity. This work provides essential guidelines for tailoring PLA properties by controlling thermal history to optimize flexibility and processability for advanced applications, specifically in melt-processing for flexible packaging. Full article

Wee1-like protein kinase 2 (WEE2) is an oocyte-specific kinase that regulates meiotic arrest and fertilization. Its largely restricted expression in female germ cells and absence in somatic tissues make it a highly selective target for reproductive health interventions. Despite its central role in human fertility, no clinically approved WEE2 modulator is available. In this study, we employed an integrated in silico approach that combines structure-based virtual screening, molecular dynamics (MD) simulations, and MM-PBSA free-energy calculations to identify repurposed drug candidates with potential WEE2 inhibitory activity. Screening of ~3800 DrugBank compounds against the WEE2 catalytic domain yielded ten high-affinity hits, from which Midostaurin and Nilotinib emerged as the most mechanistically relevant based on kinase-targeting properties and pharmacological profiles. Docking analyses revealed strong binding affinities (−11.5 and −11.3 kcal/mol) and interaction fingerprints highly similar to the reference inhibitor MK1775, including key contacts with hinge-region residues Val220, Tyr291, and Cys292. All-atom MD simulations for 300 ns demonstrated that both compounds induce stable protein–ligand complexes with minimal conformational drift, decreased residual flexibility, preserved compactness, and stable intramolecular hydrogen-bond networks. Principal component and free-energy landscape analyses further indicate restricted conformational sampling of WEE2 upon ligand binding, supporting ligand-induced stabilization of the catalytic domain. MM-PBSA calculations confirmed favorable binding free energies for Midostaurin (−18.78 ± 2.23 kJ/mol) and Nilotinib (−17.47 ± 2.95 kJ/mol), exceeding that of MK1775. To increase the translational prioritization of candidate hits, we place our structure-based pipeline in the context of modern machine learning (ML) and deep learning (DL)-enabled virtual screening workflows. ML/DL rescoring and graph-based molecular property predictors can rapidly re-rank docking hits and estimate absorption, distribution, metabolism, excretion, and toxicity (ADMET) liabilities before in vitro evaluation. Full article

The end-to-end efficiency of radio-frequency (RF)-powered wireless communication networks (WPCNs) in post-disaster underground mine environments can be enhanced through adaptive beamforming. The primary challenges in such scenarios include (i) identifying the most energy-constrained nodes, i.e., nodes with the lowest residual energy to prevent the loss of tracking and localization functionality; (ii) avoiding reliance on the computationally intensive channel state information (CSI) acquisition process; and (iii) ensuring long-range RF wireless power transfer (LoRa-RFWPT). To address these issues, this paper introduces an adaptive and safety-aware deep reinforcement learning (DRL) framework for energy beamforming in LoRa-enabled underground disaster networks. Specifically, we develop a Safe Adaptive Deep Q-Network (SADQN) that incorporates residual energy awareness to enhance energy harvesting under mobility, while also formulating a SADQN approach with dual-variable updates to mitigate constraint violations associated with fairness, minimum energy thresholds, duty cycle, and uplink utilization. A mathematical model is proposed to capture the dynamics of post-disaster underground mine environments, and the problem is formulated as a constrained Markov decision process (CMDP). To address the inherent NP hardness of this constrained reinforcement learning (CRL) formulation, we employ a Lagrangian relaxation technique to reduce complexity and derive near-optimal solutions. Comprehensive simulation results demonstrate that SADQN significantly outperforms all baseline algorithms: increasing cumulative harvested energy by approximately 11% versus DQN, 15% versus Safe-DQN, and 40% versus PSO, and achieving substantial gains over random beamforming and non-beamforming approaches. The proposed SADQN framework maintains fairness indices above 0.90, converges 27% faster than Safe-DQN and 43% faster than standard DQN in terms of episodes, and demonstrates superior stability, with 33% lower performance variance than Safe-DQN and 66% lower than DQN after convergence, making it particularly suitable for safety-critical underground mining disaster scenarios where reliable energy delivery and operational stability are paramount. Full article