Maintaining frequency stability in modern multi-area interconnected power systems has become increasingly challenging due to the stochastic nature of wind power and reduced effective system inertia. Under these dynamic conditions, traditional fixed-gain PID controllers frequently fail to provide robust regulation. To address this limitation, this study proposes and evaluates a practical model-free secondary control strategy for multi-area Load Frequency Control (LFC). The proposed hybrid MFAC–PID framework integrates an incremental model-free adaptive control (MFAC) law with a low-gain incremental PID damping term. This combination leverages real-time input–output data to determine primary control actions without relying on an explicit plant model, while the PID component supplies supplementary damping based on recent control errors. Furthermore, the controller utilizes online pseudo-gradient estimation to dynamically adapt to stochastic wind fluctuations and ±5% parametric uncertainty. Simulation results demonstrate that the hybrid design substantially enhances Area Control Error (ACE) regulation. Under wind-disturbed conditions, it reduces the aggregated Integral Absolute Error (IAEtotal) from 92.76 to 41.10, representing an improvement of over 50% compared with the fixed-gain PID baseline. Additionally, the controller maintains a low computational overhead of 0.306 milliseconds per control cycle. These findings indicate that the hybrid MFAC–PID structure provides a robust, computationally efficient solution for real-time Automatic Generation Control (AGC) in renewable-integrated multi-area power grids.

This study investigates the predictability of Environmental, Social, and Governance (ESG) performance using financial fundamentals within the context of Taiwan, a prominent small open economy integrated into global value chains. As global markets transition toward mandatory sustainability reporting, identifying the financial ante-cedents of ESG outcomes is critical for risk management and regulatory oversight. Uti-lizing a decade of firm-level data (2014–2023) from the Taiwan Economic Journal (TEJ), we employ supervised machine learning (ML) architectures-including Decision Tree, Random Forest, and Extreme Gradient Boosting (XGBoost)-to classify firms into ESG performance tiers based on indicators such as profitability, valuation, and scale. Our empirical results provide robust support for the Slack Resources Hypothesis, identifying Return on Assets (ROA) and Firm Size (SIZE) as the most consistent predictors of ESG excellence across the semiconductor, cement, and steel sectors. Conversely, mar-ket-based indicators (Tobin’s Q) dominate predictive models for the financial industry. Methodologically, XGBoost delivers superior predictive calibration for the financial sector, while Decision Trees offer highly interpretable threshold-based logic for risk screening. Our study contributes a transparent “early-warning” framework, enabling investors and regulators to identify sustainability risks through auditable financial benchmarks. The findings suggest that while financial latitude is a structural prerequisite for ESG engagement, it is not its sole determinant, pointing toward a “virtuous circle” of financial health and managerial quality.

In modern intelligent manufacturing, spindle thermal errors are critical to machining accuracy. To address this, we propose a two-stage prediction framework. First, for feature selection, an enhanced Red-Billed Magpie Optimization algorithm (RBMO-X) optimizes the parameters of a hybrid convolutional neural network (DLTK). Concurrently, PSO-optimized HDBSCAN clustering combined with Pearson correlation selects optimal temperature-sensitive points. The DLTK network integrates LSTM, deformable convolution, Transformer, and Fourier KAN modules for robust spatiotemporal feature extraction. The experimental results demonstrate significant improvements. The proposed feature selection method improves the Silhouette index by 32.39% and increases BWP by 49.16%. Using the selected points reduces prediction RMSE by 31.89% compared to random selection. The final RBMO-X-DLTK model achieves an RMSE of 0.181 μm, an MAE of 0.128 μm, and an R2 score of 0.9978, outperforming seven benchmark models (e.g., BP, LSTM, CNN-LSTM). In practical validation, the model enabled an average thermal error reduction of 89%. This integrated approach provides a robust and accurate solution for spindle thermal error prediction, demonstrating strong generalization capability.

Foundation models (FMs) have become a paradigm shift in the field of artificial intelligence, allowing one large-scale pretrained model to be customized for a broad set of downstream tasks using very little task-specific data. These models, which include GPT, CLIP, BERT, and vision transformers, have altered the scope of transfer learning and multimodal understanding and are built on top of enormous datasets and self-supervised learning. The paper provides a broad view of the modern state of foundation models, with an emphasis on their technological foundation, training, and cross-domain use in fields like natural language processing, computer vision, healthcare, robotics and scientific discovery. We also explore the main opportunities that FMs offer, as well as state-of-the-art methods and techniques for the development of foundation models. we discuss their applications in natural language processing, computer vision, healthcare, etc. Furthermore, their limitations and challenges are also investigated. Lastly, future prospects are discussed so that professionals and scientists obtain a better understanding of the importance of foundation models for addressing their research goals.

This study focuses on estimating the nine parameters of the three-diode model (3DM) for photovoltaic (PV) cells by integrating the Atan-Sinc Optimization Algorithm (ASOA) with the Newton–Raphson (NR) method. The ASOA, a population-based metaheuristic approach inspired by the behaviors of the Sech and Tanh functions, systematically generates candidate solutions for the complete set of parameters in the 3DM. For each of these solutions, the NR method is employed to solve the transcendental equation governing the solar cell model, facilitating a precise evaluation of the associated objective function. To guide the parameter estimation process, experimental current-voltage (I-V) and voltage-power (V-P) curves are utilized. The robustness of the proposed methodology is validated through studies on both monocrystalline and polycrystalline solar cells. Computational results reveal that the ASOA effectively navigates the parameter space, while the NR method provides accurate evaluations, resulting in reliable and precise parameter estimations. All numerical validations were conducted using MATLAB software, version 2024b.

Drowning remains the third leading cause of accidental injury-related deaths worldwide, disproportionately affecting low- and middle-income countries where lifeguard coverage is limited or absent. To address this critical gap, we present Robobuoy, an intelligent real-time rescue system that integrates deep learning-based object detection with an unmanned surface vehicle (USV) for autonomous intervention. The system employs a monitoring station equipped with two specialized object detection models: YOLO12m for recognizing drowning individuals and YOLOv5m for tracking the USV. These models were selected for their balance of accuracy, efficiency, and compatibility with resource-constrained edge devices. A geometric navigation algorithm calculates heading directions from visual detections and guides the USV toward the victim. Experimental evaluations on a combined open-source and custom dataset demonstrated strong performance, with YOLO12m achieving an mAP@0.5 of 0.9284 for drowning detection and YOLOv5m achieving an mAP@0.5 of 0.9848 for USV detection. Hardware validation in a controlled water pool confirmed successful target-reaching behavior in all nine trials, achieving a positioning error within 1 m, with traversal times ranging from 11 to 23 s. By combining state-of-the-art computer vision and low-cost autonomous robotics, Robobuoy offers an affordable and low-latency prototype to enhance water safety in unsupervised aquatic environments, particularly in regions where conventional lifeguard surveillance is impractical.

The flourishing of the Internet of Things (IoT) has not only improved our lives in smart homes and healthcare but also made us more susceptible to cyberattacks. Legacy intrusion detection systems are simply overwhelmed by the scale and diversity of IoT traffic, which is why there is a need for more intelligent forensic solutions. In this paper, we present a statistical technique, the Averaging Detection Method (ADM), for detecting attack traffic. Furthermore, the five deep learning models SimpleRNN, LSTM, GRU, BLSTM, and BGRU are compared for malicious traffic detection in IoT network forensics. A smart home dataset with a simulated DoS attack was used for performance analysis of accuracy, precision, recall, F1-score, and training time. The results indicate that all models achieve high accuracy, above 97%. BiGRU achieves the best performance, 99% accuracy, precision, recall, and F1-score, at the cost of high training time. GRU achieves perfect precision and recall (100%) with faster training, which can be considered for resource-constrained scenarios. SimpleRNN trains faster with comparable accuracy, while LSTMs and their bidirectional counterparts are better at capturing long-term dependencies but are computationally more expensive. In summary, deep learning, especially BiGRU and GRU, holds great promise for boosting IoT forensic investigation by enabling real-time DoS detection and reliable evidence collection. Meanwhile, the proposed ADM is simpler and more efficient at classifying DoS traffic than deep learning models.

Swarm unmanned surface vehicles (USVs) have been increasingly explored for maritime defense and security operations, particularly in scenarios requiring the rapid detection and interception of multiple attackers. The target detection reliability and defender–target assignment stability are significantly crucial to ensure quick responses and prevent mission failure. A key challenge in such missions lies in the assignment of targets among multiple defenders, where frequent reassignment can cause instability and inefficiency. This paper proposes a novel ETA-hysteresis-guided reinforcement learning (RL) framework for continuous multi-target hunting with swarm USVs. The approach integrates estimated time of arrival (ETA)-based task allocation with a dual-threshold hysteresis mechanism to balance responsiveness and stability in multi-target assignments. The ETA module provides an efficient criterion for selecting the most suitable defender–target pair, while hysteresis prevents oscillatory reassignments triggered by marginal changes in ETA values. The framework is trained and evaluated in a 3D-simulated water environment with multiple continuous targets under static and dynamic water environments. Experimental results demonstrate that the proposed method achieves substantial measurable improvements compared to basic MAPPO and MAPPO-LSTM, including faster convergence speed (+20–30%), higher interception rates (improvement of +9.5% to +20.9%), and reduced mean time-to-capture (by 9.4–19.0%), while maintaining competitive path smoothness and energy efficiency. The findings highlight the potential of integrating time-aware assignment strategies with reinforcement learning to enable robust, scalable, and stable swarm USV operations for maritime security applications.

Ensuring reliable inspection and quality control in complex industrial settings remains a significant challenge, particularly when traditional manual methods are applied to dynamic, multi-object environments. This paper presents and validates a new hybrid framework that integrates Object-Centric Process Mining (OCPM) with Support Vector Machines (SVMs) to improve industrial safety and quality assurance. The aims are: (1) to uncover and model the complex, multi-object processes characteristic of modern manufacturing using OCPM; (2) to assess these models in terms of conformance, performance, and the detection of bottlenecks; and (3) to design and embed a predictive layer based on Support Vector Regression (SVR) to anticipate process outcomes and support proactive control.The proposed methodology comprises a comprehensive pipeline: data fusion and OCEL structuring, OCPM for process discovery and conformance analysis, feature engineering, SVR for predictive modeling, and a multi-objective optimization layer. By applying this framework to a timber sawmill dataset, the study successfully modeled complex lumber drying operations, identified key object interactions, achieving a process conformance fitness score of 0.6905, and testing the integration of a predictive SVR layer. The SVR model’s predictive accuracy for production yield was found to be limited (R2=0.0255) with the current feature set, highlighting the challenges of predictive modeling in this complex, multi-object domain. Despite this predictive limitation, the multi-objective optimization effectively balanced defect rates, energy consumption, and process delays, yielding a mean objective function value of 0.0768. These findings demonstrate the framework’s capability to provide deep, object-centric process insights and support data-driven decision-making for operational improvements in Industry 4.0. Future research will focus on improving predictive model performance through advanced feature engineering and exploring diverse machine learning techniques.

Estimating the absolute scale of reconstructed camera trajectories in monocular odometry is a challenging task due to the inherent scale ambiguity in any monocular vision system. One promising solution is to fuse data from different sensors, which can improve the accuracy and precision of scale estimation. However, this approach often requires additional effort in sensor design and data processing. In this paper, we propose a novel method for fusing single-camera data with wheel odometer readings using a fuzzy system. The architecture of the fuzzy system has as inputs the wheel odometer value and the translation and rotation obtained from ORB-SLAM2. It was trained with the ANFIS tool in MATLAB 2014b. Our approach yields significantly better results compared to state-of-the-art pure monocular systems. In our experiments, the average error relative to GPS measurements was only four percent. A key advantage of this method is the elimination of the sensor calibration step, allowing for straightforward data fusion without a substantial increase in data processing demands.