The rapid progress of text-to-speech (TTS) and voice conversion (VC) technologies has enabled the creation of synthetic voices that closely mimic real speakers. While these systems have beneficial applications, they also raise security concerns, being susceptible to misuse by fraud, impersonation, and disinformation. In this paper, we propose a synthetic speech detection system based on conventional convolutional neural network (CNN) models trained from scratch on cepstral-based and constant-Q cepstral coefficients (MFCCs, LFCCs, CQCCs) as inputs. To improve robustness and reduce overfitting, we employ targeted augmentation methods (Mixup, SpecAugment) alongside regularization techniques including batch normalization, dropout, label smoothing, and adaptive learning rate scheduling. Our experiments are carried out on the Fake-or-Real (FoR) corpus, a large-scale benchmark that includes both authentic human recordings and outputs from a variety of TTS and VC systems. Our best system configuration achieves 99.4% validation accuracy and 95.9% test accuracy, outperforming our previous work and demonstrating that carefully designed CNN architectures combined with focused augmentation and regularization can deliver high detection performance.
This paper addresses the complexity of power grid operation behavior by developing a multi-scenario preview and dynamic verification method leveraging virtual reality (VR) technology. The method integrates real-time positioning data, operation ticket information, and environmental variables to construct an immersive 3D virtual scene that closely replicates actual grid equipment and conditions. Within this virtual environment, the behavior and responses of grid operators under various fault scenarios are simulated, enabling early risk identification and assessment. Dynamic verification algorithms are applied to validate operational behavior in real time, providing data-driven decision support for dispatchers. Experimental results demonstrate that the VR-based approach achieves a success rate of at least 85% in scenarios including line faults, equipment faults, load overload, and frequency fluctuations, with an average response time of 16.5 seconds and a behavior simulation recognition accuracy of about 90%. Additionally, the dynamic verification algorithm recorded a MAE (Mean Absolute Error) of 0.03 and an MSE (Mean Squared Error) of 0.002 within 60 seconds. These findings highlight the innovative integration of immersive VR and dynamic verification techniques, offering a robust tool for enhancing the safety and reliability of power grid operations.
Power electronic devices have gradually become the core components of modern energy systems. However, due to their nonlinear dynamic characteristics and operational safety of the system. Therefore, a stability optimization method for power Optimization (IIW-PSO) is proposed. This method utilizes the IIW-PSO to optimize controller parameters, thereby reducing dependence on accurate mathematical models, and combines active damping compensation to adjust damping strength in real time. The voltage tracking error was the lowest at 1.25%, and the Total Harmonic Distortion (THD) was the lowest at 3.3%. In the practical application of stability testing for power electronic equipment, the voltage surge of the improved power electronic equipment system was only 2.9%, the highest steady-state error was only 2.1, and the maximum dynamic disturbance of the output voltage was 20V. The stability method based on the IIW-PSO has excellent anti-interference performance, can meet the high-performance and high reliability requirements, and further promote the development of power electronic equipment towards intelligence and high adaptability.
This paper provides a detailed analysis of the wave energy potential at three locations in the Black Sea, evaluating the performance of various WEC technologies. The paper highlights the difficulties of implementing WEC as well as a series of financial data on the electricity produced. In this regard, the paper may be useful for operators in coastal areas to assess the possibilities of practical use of wave energy.
In recent days, there has been a significant interest in using Machine Learning (ML) techniques in different domains that serves people's lives. Interpolation and extrapolation are fundamental techniques in data prediction and ML. In contrast, interpolation estimates values within a known data range, whereas extrapolation attempts to predict beyond the observed data points. This study compares the accuracy and limitations of these approaches using synthetic datasets modeled on sinusoidal functions with noise. We employ cubic interpolation and polynomial regression to assess their predictive reliability. Results show that interpolation provides highly accurate estimates within the given range, whereas extrapolation introduces significant deviations as it moves further from known data. Our findings highlight the risks of relying on extrapolation in ML applications and emphasize the importance of selecting appropriate methods based on data distribution and predictive constraints. To further support the analysis, we implemented linear regression and a Moving Average model as a non-parametric alternative to Random Forest. Performance was evaluated using Root Mean Square Error (RMSE) and Mean Absolute Error (MAE), both numerically and analytically. Polynomial regression of degree 5 proved most effective for interpolation, while cubic splines showed instability in extrapolation. These insights reinforce the need for tailored model selection in predictive modelling tasks.
This study investigates the impact of Artificial Intelligence integration in educational environments on academic performance and competency development. Empirical data demonstrated key differences between traditional methodologies and those based on artificial intelligence. Statistical analysis identified critical implementation variables including resource allocation constraints and technological accessibility parameters. The findings demonstrate that optimal integration of artificial intelligence in educational frameworks requires a systematic approach to maintain pedagogical integrity. This research provides quantifiable metrics in combination with a variety of tests, and statistical analysis, which results in offering a proper foundation for establishing further investigations into educational technology integration and workforce preparation optimization.
With the increasing complexity of distributed energy access and operation problems of uneven sample distribution and low computational efficiency in risk assessment, making it difficult to accurately quantify the risks under the coupling of multidimensional uncertainty factors. In response to this challenge, a power grid operation risk assessment method based on improved hypercube sampling is proposed. Latin Hypercube Sampling (LHS) is used to achieve hierarchical uniform sampling, and particle swarm optimization algorithm is introduced to dynamically optimize the sample size in high-risk areas. A Monte Carlo Simulation-based Grid Operation Risk Assessment Model (MCS-GORAM) is constructed for power grid operation risk assessment. The results show that the improved method improves sample uniformity (Sobol's index 0.92) by 48.4% compared to traditional MCS, reduces convergence iteration times by 45.8%, reduces computation time by 49.4%, and reduces risk index error by only 1.23%, which is 61.7% lower than MCS. In a 500 node microgrid scenario, the expected loss of load for MCS-GORAM is 11.56 MW & centerdot;h, with a prediction accuracy improvement of more than 10% compared to the comparison model. Additionally, the standard deviation of equipment failure probability estimation is reduced to 0.006, demonstrating strong robustness to high-dimensional uncertainty. Crucially, its practical applicability was further confirmed in a simulated real-world industrial microgrid case study, where it consistently maintained the highest accuracy and efficiency. The research provides an efficient quantitative tool for risk assessment of power grids containing new energy, which has important engineering application value for ensuring the safe operation of power grids and optimizing resource allocation.
This study presents an analysis based on the dispersion of atmospheric pollutants resulting from a cogeneration power plant, using the Gaussian model. The objective of the research is to evaluate the impact generated on the environment as a result from energy production. The study integrates the emission source and meteorological parameters, as well as dispersion coefficients, to determine the amount of pollutant concentrations recorded at local level. The simulation shows that the maximum values of concentration for analyzed pollutants (NOx, SO2, CO, VOCs) levels are still generally below the ambient air quality standard values set by Law No. 104/2011, suggesting a minimal background pollution contribution.
Vegetation indices are useful tools in smart agriculture for many applications such as assessment ofplant health, estimation ofyield, drought prediction. This study examines the measure in which each of these specialized vegetation indices contributes to crop discrimination in a classification process. We begin with a correlation study of 36 vegetation indices computed on PRISMA hyperspectral images. Then we propose an algorithm based on principal components for the removal of redundant vegetation indices, while the efficiency is measured over the resulting images with structural similarity. Finally, a comparison is made in terms of segmentation results using Random Forests, ResNet, U-Net and SE-ResNet on the whole hyperspectral image vs. a multiband selected vegetation indices image. Our results show that the proposed vegetation index selection method is effective for hyperspectral image segmentation
To address the inefficiency and safety risks in substation equipment inspection, a multi-sensor fusion-based detection method is proposed. We extend YOLOv8 model by integrating the CBAM (Convolutional Block Attention Module) to boost the capacity for feature extraction, while the ATFL (Adaptive Threshold Focal Loss) loss function is adopted to mitigate complex background interference. Compared to the baseline YOLOv8, the improved model achieves increases of 1.9% in precision, 3.5% in recall, and 2.2% in mAP. For point cloud processing, multi-sensor joint calibration enables image-point cloud fusion. After preprocessing with voxel and pass-through filtering, Euclidean clustering is applied to segment substation equipment point clouds within detection regions. The results obtained from experiments reveal that the proposed method achieves accurate equipment identification and reliable pose estimation, meeting practical inspection requirements.
When a converter transformer with residual flux in its iron core undergoes energization, it may generate significant magnetizing inrush current while simultaneously inducing sympathetic inrush current in parallel-connected transformers due to magnetic coupling effects. The substantial harmonic components contained within these inrush currents can severely distort the busbar voltage waveform, resulting in temporary voltage depression and potentially triggering erroneous operations in wideband harmonic protection systems. To systematically address this critical issue, this paper conducts an in-depth investigation into the underlying physical mechanisms and system-wide impacts of complex inrush current phenomena in converter transformers, followed by a comprehensive analysis of wideband harmonic protection operational principles and key influencing factors affecting its performance during transient inrush conditions .An optimized wideband harmonic protection scheme based on a dynamic protection threshold is proposed. This method employs a real-time adaptive threshold that varies with the system operating state to mitigate protection maloperation caused by inrush current during operational transitions in HVDC transmission systems. A bipolar HVDC system simulation model is established to replicate the inrush current process during converter transformer energization, validating the effectiveness of the proposed method in practical engineering applications. Simulation results demonstrate that, compared to conventional fixed thresholds, the dynamic threshold-based protection algorithm significantly reduces the maloperation rate without introducing new failure scenarios, thereby enhancing the stability and reliability of the power system.
Image style transfer has been widely applied in the field of computer vision, primarily aiming to transform captured images into specific artistic styles using computational techniques. Traditional algorithms, such as those based on structural tensors and ETF (Edge Tangent Flow) or combining Kuwahara filtering with line integral convolution, rely heavily on heuristic construction by researchers' domain knowledge or experience. Most style transfer methods depend on sample-based training, resulting in high computational costs and slow processing speeds. To address these limitations, this study proposes an image line drawing generation algorithm under direction field consistency constraints. The proposed approach preprocesses images using non-local means filtering, computes the direction field through Tikhonov regularization accelerated by the Sherman-Morrison-Woodbury formula, and guides DoG (Difference of Gaussians) filtering with the regularized direction field. A perception threshold algorithm, designed based on human visual system characteristics, binarizes the filtered results to generate line drawings. Simulation experiments demonstrate that the algorithm effectively transforms input images into line drawings with smooth lines and preserves the primary information of the original images while significantly improving computational efficiency.
This paper constructs an interactive active distribution network architecture to get a real-time power system state. It uses a multi-time scale coordinated optimization process and builds physical models for new loads. Applying a genetic algorithm solves the system's optimal state, obtaining the global optimal solution and reducing shear load and wind and light abandonment. It cuts power system costs, boosts the economy, adapts to diverse loads, and expands scenarios.
To address challenges in improving renewable energy utilization efficiency and mitigating node overvoltage issues caused by high-penetration distributed photovoltaic (PV) integration in low-voltage distribution networks (LVDNs), while enhancing voltage security margins, this paper proposes a multi-objective optimization model and power management strategy for LVDNs integrated with distributed PV and distributed energy storage systems (DESS). Four comprehensive simulation cases validate the proposed approach. Results demonstrate that this method effectively maintains node voltages within permissible limits while ensuring the economic operation of LVDNs.
Fluctuations of critical coupling coefficient in parity-time (PT) symmetric wireless power transfer (WPT) systems have significant influences on their transfer performance. To solve this problem, this research proposed a method based on power electronic topology to stabilize its transfer performance. Firstly, a mathematical model was established based on two-coil magnetically coupled resonant wireless power transfer (MCR-WPT) topology of the series-series (S-S) type in a stable state. The formulation of the critical coupling coefficient of the system was derived considering stable state. Then, a control approach based on output-end Buck-Boost converter was designed, which could be applied for the stabilization of the critical coupling coefficient of the system. This approach ensured that the system could provide a stable PT-symmetric solution space under variable loads. Finally, a WPT system was designed, boasting transfer efficiency of 80% and output power of 40W. The transfer characteristics of the system under load perturbation was verified using simulation and experiment. Research results demonstrated that the proposed method could retain the system with constant critical coupling coefficient and stable output power and transmission efficiency under load disturbance.
This study introduces a novel Dual-Path Learning Framework (DPLF) that synergistically integrates Mean Teacher-based semi-supervised learning with contrastive regularization to address the significant annotation costs associated with medical imaging (e.g., approximately 8-12 hours per pathological case, incurring expenses exceeding USD 70 per 3D case). Building upon the traditional Mean Teacher paradigm, DPLF incorporates three key innovations: a boundary-aware contrastive loss function, a dynamic weight allocation strategy, and a dual-path network architecture. These enhancements collectively improve segmentation accuracy and model robustness, particularly in scenarios with limited labeled data. Notably, the contrastive regularization module reduces false positives in vascular boundary segmentation by 18.7%, while the boundary segmentation error decreases significantly from 7.7 to 5.2 pixels (p < 0.01). Remarkably, when trained with only 10% of the labeled data, DPLF achieves 92.1% of the fully supervised model's performance, substantially outperforming the baseline method's 76.3%. By embedding boundary-aware similarity constraints into feature learning, DPLF effectively preserves critical anatomical structures and delivers superior performance in downstream applications such as histopathological image analysis and radiomics feature extraction. This framework offers an efficient, accurate, and clinically viable solution for advanced medical imaging analysis.
Urban traffic management requires rapid evaluation of control strategies, yet traditional sequential simulations are too slow for real-time decision support. This paper presents a scalable Digital Twin framework that leverages parallel microscopic simulations in SUMO to assess alternative speed settings in congested areas. The system compares throughput, delay, and simulation-to-real-time ratios, enabling optimal speed recommendations within strict time constraints. Results show that anticipatory simulations can run faster than real-time on commodity hardware, demonstrating the potential of parallel digital twins to provide timely, data-driven support for adaptive speed control in complex urban mobility systems.
This paper unveils the blueprint of a sophisticated multi-point temperature and humidity monitoring system, crafted using LabVIEW software. The article is meticulously organized into five chapters: an overview of the system's architecture, the design of the intuitive front panel, the intricacies of the program panel, the system's debugging and operational phases, and a concluding remark. The front panel is functional and aesthetic, segmented into a temperature display module, a humidity display module, and a data acquisition module that tirelessly gathers vital information. Meanwhile, the program panel is divided into a temperature module that senses the heat, a humidity module that gauges the dampness, a time module that marks the passage, and a data acquisition module that orchestrates the influx of data. The final running result shows that the system can achieve the expected effect. The system can perform real-time measurement, display, and alarm for multi-point temperature and humidity in environments such as homes and outdoors. It plays an important and positive role in people's production and daily life.
To address undetected instances and false alarms in YOLOv8s caused by small and occluded targets in power field operator detection, an improved algorithm, Sparse Depthwise Convolution and CARAFE-YOLOv8s(SCR-YOLOv8s), based on Sparse Depthwise Convolution (SPDConv) and CARAFE, is proposed. First, the Sparse Depthwise Convolution and GAM-C2f (SGAM-C2f) module is designed, replacing some backbone convolutions with non-strided convolution Sparse Depthwise Convolution (SPDConv) while introducing the Global Attention Mechanism (GAM) to enhance global contextual feature extraction for small targets. Second, the lightweight upsampling operator CARAFE replaces the original method to expand the receptive field. Third, the Repulsion-Loss function is integrated to improve detection accuracy for occluded targets in complex environments. Finally, the dynamic detection head DyHead is adopted to enhance small target processing. Experiments show SCR-YOLOv8s achieves 90.23% mAP@0.5, a 4.81% improvement over YOLOv8s, with 15.05M parameters and 32.9GFLOPs, meeting engineering application requirements.
This paper proposes an innovative security architecture for energy infrastructures, integrating advanced principles of artificial intelligence, Zero Trust, and blockchain. The key contribution of the study lies in the specific and demonstrable combination of Random Forest and K-Nearest Neighbors (KNN) machine learning algorithms for real-time anomaly detection with a Zero Trust microsegmentation architecture. Unlike previous integrated solutions, our rigorous evaluation includes a cost-benefit analysis that justifies the initial investment through a 15% reduction in long-term maintenance costs. Experimental results on a simulated SCADA system demonstrate a 40% reduction in attack detection time and a significant increase in infrastructure resilience.
