Coal Mine Accident Risk Analysis with Large Language Models and Bayesian Networks

Sustainability, Journal Year: 2025, Volume and Issue: 17(5), P. 1896 - 1896

Published: Feb. 24, 2025

Coal mining, characterized by its complex operational environment and significant management challenges, is a prototypical high-risk industry with frequent accidents. Accurate identification of the key risk factors influencing coal mine safety critical for reducing accident rates enhancing safety. Comprehensive analyses investigation reports provide invaluable insights into latent underlying mechanisms In this study, we construct an integrated research framework that synthesizes large language models, association rule Bayesian networks to systematically analyze 700 reports. First, model employed extract factors, identifying multiple layers risks, including 14 direct, 38 composite, 75 specific factors. Next, Apriori algorithm applied 281 strong rules, which serve as foundation constructing network comprising 127 nodes. Finally, sensitivity analysis path are conducted on reveal seven primary primarily related on-site management, execution procedures, insufficient supervision. The novelty our lies in efficient processing unstructured text data via significantly enhances accuracy comprehensiveness factor compared traditional methods. findings robust theoretical practical support offer valuable practices other industries. From policy perspective, recommend government strengthen legislation supervision particular focus enforcement procedures promote comprehensive education training enhance frontline personnel’s awareness emergency response capabilities, leverage data-driven technologies develop intelligent early-warning systems. These measures will improve precision efficiency scientific basis prevention control.

Language: Английский

---

A Secure and Efficient Framework for Multimodal Prediction Tasks in Cloud Computing with Sliding-Window Attention Mechanisms

Applied Sciences, Journal Year: 2025, Volume and Issue: 15(7), P. 3827 - 3827

Published: March 31, 2025

An efficient and secure computation framework based on the sliding-window attention mechanism sliding loss function was proposed to address challenges in temporal spatial feature modeling for multimodal data processing. The aims overcome limitations of traditional methods privacy protection, feature-capturing capabilities, computational efficiency. experimental results demonstrated that, time-series processing tasks, method achieved precision, recall, accuracy, F1-score values 0.95, 0.91, 0.93, respectively, significantly outperforming federated learning, multi-party computation, homomorphic encryption, TEE-based approaches. In these metrics reached 0.90, 0.92, also surpassing all comparative methods. Compared with existing frameworks, approach substantially enhanced efficiency while minimizing accuracy loss, ensuring privacy. These findings provide an reliable solution protection security cloud computing environments. Furthermore, research demonstrates significant theoretical value practical potential real-world scenarios such as financial forecasting image analysis.

Language: Английский

---

Human-Centered Digital Twins in IoT

IGI Global eBooks, Journal Year: 2025, Volume and Issue: unknown, P. 189 - 210

Published: May 2, 2025

The integration of Human-Centered Digital Twins (HCDTs) and the Internet Things (IoT) is revolutionizing industries by allowing personalized, real-time decision-making through use continuous data streams. These systems utilize IoT sensors AI-driven models to produce digital copies individuals, environments, or systems, providing improved predictive capabilities in healthcare, smart cities, industrial applications. increasing HCDTs sparks significant ethical issues, such as privacy, confidentiality, discriminatory practices, consent based on complete information. A gap persists research, particularly establishment uniform frameworks implementation dependable AI that safeguard user autonomy while optimising advantages twins. purpose this investigation investigate consequences personalization suggest a framework for reconciling data-driven with privacy cybersecurity environments.

Language: Английский

---

A Methodology Based on Deep Learning for Contact Detection in Radar Images

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(19), P. 8644 - 8644

Published: Sept. 25, 2024

Ship detection, a crucial task, relies on the traditional CFAR (Constant False Alarm Rate) algorithm. However, this algorithm is not without its limitations. Noise and clutter in radar images introduce significant variability, hampering detection of objects sea surface. The algorithm’s theoretically Constant Rates are upheld practice, particularly when conditions change abruptly, such as with Beaufort wind strength. Moreover, high computational cost signal processing adversely affects process’s efficiency. In previous work, four-stage methodology was designed: first preprocessing stage consisted image enhancement by applying convolutions. Labeling training were performed second using Faster R-CNN architecture. third stage, model tuning accomplished adjusting weight initialization optimizer hyperparameters. Finally, object filtering to retrieve only persistent objects. This work focuses designing specific for ship Peruvian coast commercial images. We two key improvements: automatic cropping labeling interface. Using artificial intelligence techniques leads more precise edge extraction, improving accuracy cropping. On other hand, developed interface facilitates comparative analysis persistence three consecutive rounds, significantly reducing times. These enhancements increase efficiency enhance learning model. A dataset consisting 60 used experiments. Two classes considered, cross-validation applied validation models. results yield value 0.0372 function, recovery rate 94.5%, an 95.1%, respectively. demonstrates that proposed can generate high-performance contact

Language: Английский

---