Power transformers are among the most costly assets in power networks, of which differential protection is one of their main protections. A major challenge in differential protection is discriminating internal and winding faults from inrush current (IC), sympathetic inrush current (SIC), overexcitation (OE), and external faults, with and without IC, SIC, and OE. Various methods have been proposed for this purpose; however, they are not necessarily simple to implement and often do not cover all disturbances. In this paper, a disturbance detection approach based on the fast discrete orthonormal Stockwell transform (FDOST) of the differential current is proposed. The average, kurtosis, and relative maximum of the absolute value of the FDOST coefficients are considered as discriminative features. Advantages of the approach include simple implementation and a low computational burden. The results demonstrate the satisfactory performance of the approach in achieving an accuracy of over 99% with an average processing time of 0.2 ms and a detection time of approximately 13 ms. Furthermore, the results are compared with those of the well-known second harmonic restraint method, verifying that the proposed method achieves higher accuracy with a shorter detection time. Moreover, the approach maintains robust performance under noisy conditions, current transformer saturation, and variations in sampling frequency. Additionally, experimental results obtained from a laboratory transformer confirm the high accuracy of the approach in discriminating internal faults from ICs.
Recommender systems have become a fundamental component of modern online platforms,significantly influencing user experiences by delivering personalized content. However, theincreasing complexity of contemporary recommendation models has raised critical concernsregarding their transparency and fairness. This survey provides a comprehensive and unifiedreview of explainable and fair recommender systems, emphasizing the need to move beyondaccuracy-centric design toward trustworthy and responsible recommendation paradigms. Wefirst introduce the core concepts of recommender systems, explainability, and fairness, anddiscuss their roles in building transparent and accountable systems. We then propose a multidimensional taxonomy that systematically categorizes existing approaches based on objectives,intervention stages, explanation types, and fairness perspectives. Furthermore, we present adetailed analysis of state-of-the-art methods, including explainable recommendation techniques,fairness-aware models, and emerging approaches that jointly address both aspects. In addition,we examine commonly used evaluation metrics and provide a comprehensive analysis of thetrade-offs among accuracy, explainability, and fairness, highlighting the challenges of multiobjective optimization in real-world systems. Finally, we identify key open challenges, includingdata bias, methodological limitations, evaluation inconsistencies, and deployment constraints,and outline promising directions for future research. This survey aims to provide researchers andpractitioners with a unified perspective and to support the development of recommender systemsthat are not only accurate, but also transparent, fair, and trustworthy
The growing mismatch between energy generation and consumption highlights the need for efficient energy storage solutions. Lithium-ion batteries dominate current applications due to their high energy density and long cycle life, yet their safe and optimal operation relies heavily on accurately monitoring two internal states: the state of charge (SOC) and the core temperature. Real-time estimation of these states remains difficult because the battery’s electrical and thermal behaviors are strongly coupled and vary with operating conditions. This paper presents an enhanced electro-thermal modeling and estimation framework that explicitly accounts for the mutual dependence between SOC and core temperature. Unlike conventional fixed-parameter approaches, the model parameters are defined as nonlinear functions of SOC and core temperature, capturing their inherent interdependence. These relationships are identified experimentally through controlled charge–discharge tests at 5°C, 25°C, and 45°C. A physics-based thermal model is used offline as a virtual core temperature sensor to generate reference temperature trajectories for parameter identification, since only surface and ambient temperatures are measured. The ARMAX technique is applied to estimate the electrical and thermal sub-model parameters from experimental data, and an Extended Kalman Filter is subsequently implemented on the coupled nonlinear model to jointly estimate SOC and core temperature under varying load and ambient conditions. Simulation results demonstrate that the proposed method offers higher accuracy and robustness than traditional fixed-parameter models, reducing mean estimation errors by more than 40
The ongoing opioid crisis has highlighted the urgent need for novel analgesics with low addiction potential that provide effective pain relief. In this study, we present an integrated computational framework that combines quantum chemical descriptors with classical molecular descriptors to predict the bioactivity of synthetic opioid analogs across multiple opioid receptor subtypes. Using a curated dataset of over 7000 compounds with experimentally measured activities against the delta opioid receptor (DOR), kappa opioid receptor (KOR), and mu opioid receptor (MOR), we generated several classical and quantum chemical descriptors. These high-dimensional descriptors were used to train and validate several supervised models, including random forest (RF), gradient boosting (GB) with XGBoost, support vector regression (SVR), and linear regression. The RF achieved the highest predictive performance, outperforming GB, SVR, and linear regression across all metrics. Among the descriptors, dipole moment and TPSA were the most effective, significantly enhancing model interpretability and mechanistic insight into ligand–receptor interactions. Clustering and chemical space visualization were used to identify ligand subpopulations with receptor subtype preferences, guiding rational drug design. The integration of quantum-informed descriptors, classical physicochemical features, structural clustering, and interpretable machine learning—particularly RF represents a novel contribution to computational opioid ligand design and advances the field by enabling large-scale, mechanistic analysis of opioid bioactivity. Our approach uniquely combines quantum-informed GNN descriptors with interpretable ML and classical molecular descriptors, enabling mechanistic insights beyond classical QSAR. Unlike prior works relying on empirical fingerprints or small-scale docking, this quantum–ML fusion reveals mechanistic insights into stereo electronic features driving biased agonism, offering a scalable platform for GPCR drug design amid the opioid crisis.
Accurate segmentation and classification of brain tumors from Magnetic Resonance Imaging (MRI) remain key challenges in medical image analysis, primarily due to the lack of high-quality, balanced, and diverse datasets with expert annotations. In this work, we address this gap by introducing BRISC, a dataset designed for brain tumor segmentation and classification tasks, featuring high-resolution segmentation masks. The dataset comprises 6,000 contrast-enhanced T1-weighted MRI scans, which were collated from multiple public datasets that lacked segmentation labels. Our primary contribution is the subsequent expert annotation of these images, performed by certified radiologists and physicians. It includes three major tumor types, namely glioma, meningioma, and pituitary, as well as non-tumorous cases. Each sample includes high-resolution labels and is categorized across axial, sagittal, and coronal imaging planes to facilitate robust model development and cross-view generalization. To demonstrate the utility of the dataset, we provide benchmark results for both tasks using standard deep learning models. The BRISC dataset is made publicly available.
