This paper presents an extended Power over Data Lines (PoDL) system for automotive Single-Pair Ethernet (SPE) applications. As autonomous vehicles require an increasing number of sensors and electronic modules, reducing cable weight and wiring complexity while ensuring reliable power and data transmission has become essential. Although IEEE 802.3bu defines PoDL for SPE systems, its conventional power range is limited for high-power automotive devices. The proposed system delivers up to 35 W through an SPE cable and is experimentally evaluated under various cable lengths, power sourcing equipment (PSE) supply voltages, and powered device (PD) load conditions. Voltage drop, power loss, power efficiency, packet loss, and communication latency are measured to assess system performance. Experimental results show that the proposed system achieves power efficiencies from 87.42
Accurate short‑term electricity price prediction is essential for ensuring market stability and supporting strategic energy planning. Based on real electricity market data and policy reports from the Frankfurt region of Germany, this study develops hybrid machine learning frameworks for energy price prediction. Light Gradient Boosting Regression (LGBR) and Gaussian Process Regression (GPR) are integrated with three nature-inspired optimization algorithms; Hippopotamus Optimization Algorithm (HOA), Secretary Bird Optimization Algorithm (SBOA), and Dung Beetle Optimizer (DBO); to enhance predictive performance through hyperparameter tuning. Model performance is evaluated using R², MAE, RMSE, NSE, and MAPE, while data splitting for training and testing is conducted through K-fold cross-validation to ensure robust generalization evaluation. Sensitivity analysis is performed using ANOVA to examine the influence of input variables on model performance. Among all developed models, the LGBR-HOA hybrid (LGHA) demonstrates the highest predictive accuracy in the test phase (R²=0.9933) with a runtime of 46.3 s, outperforming other hybrid and single-model configurations. The results confirm that integrating regression-based ML models with population-based optimization algorithms significantly improves forecasting accuracy and robustness in electricity price prediction.
In this paper, we propose a ray-tracing-based propagation environment analysis method for evaluating the interference effects of FM broadcast signals on aeronautical VHF radio communications in mountainous terrain. The analysis area is modeled using digital elevation model (DEM) data, and frequency-dependent radiation patterns of the FM broadcast transmitting antenna are obtained and applied in the propagation analysis to reflect terrain-dependent propagation and the radiation characteristics of the interference source. Based on this model, third-order intermodulation (IM3) interference and receiver desensitization caused by FM broadcast signals are jointly analyzed for aeronautical VHF radio communications at airborne platform altitudes of 1,000 ft, 2,000 ft, and 3,000 ft. Interference-vulnerable areas and flight segments are evaluated from the spatial received-power distribution and the power variation along the flight path. The results show that exposure to both IM3 interference and receiver desensitization is concentrated in flight segments near the FM broadcast transmitter and is stronger at 2,000 ft than at the other altitudes. These results indicate that predicting FM broadcast interference to aeronautical VHF communications in mountainous terrain requires the joint consideration of actual terrain information, transmitting antenna radiation patterns, flight altitude, IM3 interference, and receiver desensitization.
The static structure and limited control capabilities of traditional distribution networks are increasingly insufficient to meet the growing and variable energy demands of today. Therefore, there is a need for smart grid solutions that can enhance the flexibility, stability, reliability, and economic performance of distribution systems. In the literature, studies have examined distributed generation (DG) units and capacitors in conjunction with network reconfiguration (NR); however, smart grid components such as battery energy storage system (BESS) and demand response (DR) are often neglected. In this study, the optimization of DG and capacitor placement along with NR is carried out in a smart grid considering the presence of BESS and DR. The main objective of this work is to improve the technical and economic performance of an 86-bus distribution system by minimizing the total system cost and voltage deviation. The total system cost includes the installation, purchase, and operational costs of DG and capacitors, as well as the cost associated with active power losses. Optimization studies are performed under peak load conditions using Grey Wolf Optimization (GWO), Particle Swarm Optimization (PSO), and Salp Swarm Algorithm (SSA). The results indicate that the PSO algorithm provides the best objective function values across all considered scenarios. PSO based optimization of NR, DG, and capacitor placement incorporating BESS and DR leads to an approximately 82
Image–text matching is a key task in vision–language understanding, aiming to achieve fine-grained alignment between images and texts. Existing approaches primarily focus on cross-modal similarity, often failing to sufficiently capture the structural relationships among elements within each modality (i.e., intra-modal relationships), which leads to degraded performance in complex semantic alignment scenarios. To address this limitation, we propose an Entropy-based Cross-modal Relational Alignment via Linguistic Pivots (ERAL) framework, which employs Swin Transformer as the image encoder and BERT as the text encoder. ERAL explicitly models intra-modal relational structures among image regions and among text tokens by leveraging the entropy of attention distributions, and effectively incorporates these intra-modal relational cues into the cross-modal alignment process through a linguistic pivot selection mechanism, enabling more precise image–text alignment. Experimental results on the widely used Flickr30K dataset demonstrate that ERAL achieves a Recall@1 of 83.6
