ABSTRACT This paper aims to solve the problem of reducing false detection and missed detection of dynamic targets under complex backgrounds or forms and the difficulty of robots dodging obstacles in complex. Using FPN, PANet, CSPDarknet53, and CSP modules, the method improves YOLOv4-based robot vision to achieve robust dynamic target recognition in complex, occluded scenes, improve multi-scale feature fusion, and reduce computation. By combining with the GAN (Generative Adversarial Network) adversarial network, dynamic targets can be detected more accurately and effectively. The results show that the detection accuracy (mAP) is 90%, the LOSS value of the model tends to be 0, and the stability of the model is strong. The recognition prediction results for each category are as high as more than 90%, which can effectively solve false detection and missed detection. For the research on obstacle avoidance, this paper uses DWA (Dynamic Window Approach) technology for fusion and conducts multiple experiments under different conditions. The obstacle success rate reaches 95% and 90% respectively in static and dynamic environments. By improving the YOLOv4 model algorithm, it can handle dynamic target recognition and obstacle avoidance problems well, providing strong evidence for identification plates

ABSTRACT The insulating system constitutes the foundation of transformer reliability, with the insulating liquid serving as its most vital component. The progressive ageing of this liquid expedites the deterioration of insulation, thereby emphasising the imperative of continuous condition monitoring. Monitoring the key properties of transformer insulating liquids enables early detection of emerging faults and supports timely maintenance, thereby enhancing transformer reliability, safety, and service longevity. This study investigates the effects of service-induced ageing on transformer insulating liquids through a detailed analysis of critical diagnostic parameters. Evaluations performed on insulating liquid samples from in-service transformers of varying operational lifespans demonstrate a systematic and progressive degradation across all measured properties. The findings demonstrate the effectiveness of insulating liquid diagnostics as a reliable tool for assessing condition, tracking ageing progression, and enhancing the service reliability of transformers.

ABSTRACT Cheerleading is a complex competitive sport requiring high levels of strength, flexibility, coordination, and movement precision. To address the demands of hybrid sports education, this study proposes a biomechanically enhanced adaptation of the National Academy of Sports Medicine Optimum Performance Training (NASM-OPT) model for cheerleading instruction. An AI-integrated wearable system was used to provide real-time biomechanical feedback and dynamically adjust training protocols, supporting injury prevention and performance enhancement. A 12-week controlled experiment was conducted at Tianjin Sino-German University of Applied Sciences, involving an experimental group using the optimised NASM-OPT model and a control group following traditional training. The experimental group completed progressive training aligned with the five NASM-OPT phases, incorporating biomechanical factors such as joint angles, load distribution, and movement efficiency, alongside a hybrid teaching approach combining online theory and offline practice. Performance was assessed through indicators including explosive power, dynamic balance, flexibility, and coordination. Results showed significant improvements in the experimental group across all metrics (p < 0.01). These findings demonstrate that integrating biomechanical principles into structured training frameworks effectively enhances cheerleading performance in hybrid learning environments and offers a scalable approach for sports education.

ABSTRACT The rapid adoption of Virtual Reality (VR) and Augmented Reality (AR) in education has created new opportunities for enhancing English speaking and listening skills. This study examines the comparative effectiveness of VR- and AR-supported learning environments among 400 English learners at Tianjin Foreign Studies University. A mixed-method experimental design was employed, incorporating pre-tests, post-tests, and surveys administered via Google Forms. Multivariate analysis of covariance (MANCOVA) was used to compare VR, AR, and traditional instruction while controlling for prior proficiency, learner motivation, and engagement. Results showed that VR significantly improved speaking performance (F (1,395) = 466.58, p < 0.001, η² = 0.541), whereas AR led to greater gains in listening skills (F (1,395)= 396.50, p < 0.001, η² = 0.501). Estimated marginal means indicated that VR was most effective for speaking (AdjM = 90.20), while AR was superior for listening (AdjM = 89.74). No significant interaction with prior proficiency was found (p > 0.05). The findings highlight the skill-specific advantages of immersive technologies and provide practical implications for designing effective English language learning environments.

ABSTRACT Demand response (DR) has grown as an effective method for enhancing energy system flexibility and reliability, especially with the rise of renewable energy sources. However, uncertainty surrounding Distributed Energy Resources (DERs) presents challenges for Microgrid (MG) operators. Traditional optimization methods often fail to efficiently explore the vast solution space and identify optimal DR schedules within acceptable time frames. This paper proposes the Optimisation-based Scheduling of DR (OSDR) strategy, which leverages a new optimization technique, the SIAHO algorithm. Inspired by the foraging behavior of hummingbirds, the SIAHO algorithm provides a powerful framework for solving complex optimization problems. By mimicking the search patterns of hummingbirds, it efficiently explores the solution space and identifies DR schedules that balance energy costs, load factor, operation costs, and risk. Through extensive simulations and case studies, we evaluate the proposed approach across different scenarios, comparing it with existing methods. The results show that the SIAHO-based approach significantly improves energy cost reduction, peak demand shaving, and grid stability, demonstrating its robustness and effectiveness in optimizing and scheduling DR strategies.

ABSTRACT Proper cost management and performance improvement in oilfield production are deeply context-dependent and necessitate the understanding of Reservoir knowledge modelling and their impact on the operational efficiency and resource distribution. A KCMPO (Knowledge-Driven Cost Management and Performance Optimisation) framework is proposed which combines reservoir knowledge modelling, activity-based cost analysis, and data-driven decision intelligence. The use of machine learning techniques such as clustering and pattern recognition helps to discover cost-driving activities and to build dynamic links between reservoir properties, production behaviour, and monetary results. An ICT-enabled knowledge management framework that allows for the support of transparency, traceability, and continual learning across the various layers of decision-making is also included. The empirical assessment shows that the developed framework increases both cost efficiency and productive performance, thus establishing a data-supported decision-making process in oilfields that is systematic and sustainable.

ABSTRACT The integration of Collaborative Learning Technologies (CLT) and Human–Computer Interaction (HCI) in college English teaching has gained attention for enhancing interactivity, collaboration, and learning outcomes. This study evaluates the impact of HCI-based tools and CLT platforms on student participation, collaborative behaviors, and performance compared to traditional instruction. A 12-week experimental study was conducted with 500 undergraduate students, evenly divided into a control group (n = 250) receiving traditional teaching and an experimental group (n=250) using HCI tracking tools and collaborative platforms such as Google Docs, Moodle, and Zoom. Data were collected through pre- and post-tests, engagement metrics, student surveys, and teacher feedback. Quantitative analyses (t-tests, ANOVA) and qualitative thematic analyses were conducted. Results showed that the experimental group outperformed the control group with 12.3% higher post-test scores and demonstrated 17% and 14% greater engagement in interaction frequency and session duration, respectively. Survey responses indicated that 82% of students reported increased engagement, and 76% reported improved collaboration skills. These findings support the effectiveness of integrating HCI and CLT in enhancing English language learning outcomes and promoting collaborative, technology-driven educational practices.

ABSTRACT In the context of digital education, this study proposes a data-mining–driven approach to improve key knowledge analysis in English teaching. A small-sample relational inductive reasoning model is developed by integrating graph neural networks (GNNs) with meta-learning to address data scarcity and personalised learning demands. An English teaching knowledge graph is constructed using digital textbooks, lesson plans, and authoritative language knowledge bases, with entity and relation extraction techniques used to model semantic relationships. To enhance inductive reasoning under limited data conditions, path masking and loss-weight meta-learning strategies are introduced. Experimental results show that the proposed model imports tens of millions of nodes in 38 seconds, achieving an 82% efficiency improvement over the basic GNN model. The multi-hop query delay is reduced to 6 ms, 98% lower than that of traditional databases, with significantly optimised memory usage. In practical teaching applications, students’ average reading scores increase by 8.2 points, grammar mastery reaches 92%, personalised learning path matching achieves 88%, and the repetition rate of incorrect answers decreases by 72%. Overall, the proposed method substantially enhances English teaching effectiveness and learning outcomes.

ABSTRACT To explore the safety factors of new energy trackless rubber wheels in coal mines, this paper is based on the TOPSIS model, combined with the grey association level algorithm for experimental analysis, and deeply analyzes the safety-influencing factors of new energy trackless rubber wheels in coal mines. The results show that, between 2016 and 2021, the safety index of new energy trackless in coal mines declined slightly; between 2022 and 2023, Its value increased from 0.4493 to 0.5496; there was a large increase; this proves that the safety impact level of the new energy trackless rubber wheel car in the coal mine is in the ‘critical safety’; From 2024 to 2030, the application safety value of new energy trackless rubber wheels in coal mines will be maintained between 0.4628 and 0.5016, there is a negative effect, this proves that the impact of new energy safety factors on the development of new energy has a negative correlation, the hierarchical analysis method has a practical effect on the safety factor analysis of new energy trackless rubber wheels in coal mine.

ABSTRACT While the presented setting involves decentralised training of an ML model for privacy-preserving analytics in healthcare, the coupling of the two has also started to usher in significant security loopholes, predominantly relating to secure model aggregation while keeping key management efficient for resource-constrained IoMT devices. Hence, this paper aims to implement a hybrid bio-inspired optimisation method for key generation in a novel secure data storage framework. At the heart of the solution is the STBFO algorithm, which combines the best properties of two different natural-inspired techniques to generate cryptographic keys of very high entropy and strength. Hence, the architecture is designed to accommodate the Noise Parameter Server with the multi-phase key management scheme for sound privacy preservation. In testing the framework against real-life clinical datasets, both structured patient records and medical image data were used. The method brought about 17% less encryption time and 16.9% faster decryption compared to existing schemes. Having a 95.3% rate of privacy preservation and a 98.4% rate of success in authentication, our system is actually proven to be an efficiently secure anonymous user-preserving system for FL-enabled IoMT systems.