Significant challenges are presented by the widespread problem of harmonic distortions in power systems, which result from the non-linearity in electrical power distribution. Inaccurate power metering, motor and capacitor failures, equipment overheating, and other manifestations of these distortions underscore the importance of developing reliable solutions. This paper presents Runge-Kutta-driven computational intelligence to effectively solve the harmonics estimation problem. An objective function specific to power system inter and sub-harmonics is developed to simultaneously estimate phase and amplitude parameters. The Runge-Kutta optimization algorithm (RKOA) demonstrates its effectiveness in precisely estimating phase and amplitude parameters across various settings, achieving an impressive precision of 5.31E-09 in harmonic signal estimation. Three different harmonic signal Examples, including inter-harmonics and sub-harmonics, are carefully investigated in this work. The RKOA yields mean squared error (MSE) values of 6.18E-07, 7.07E-09, and 5.31E-09 for Example 1 with particle sizes ranging from 50 to 250 and noise levels of 80 dB, 40 dB, and 20 dB, respectively. Example 2 achieves MSE values as low as 6.07E-09 when more complicated inter and sub-harmonics are included. For the dynamic harmonic case in Example 3, performance is evaluated through time-domain signal reconstruction, demonstrating accurate tracking of time-varying amplitudes under non-stationary conditions. The results show a step forward in the development of an accurate and robust computational framework for reliable harmonic estimation in contemporary power and energy systems.
Virtual reality (VR) can foster self-directed learning (SDL), yet traditional feedback in self-directed VR (SDVR) environments often fails to provide timely and individualized support, particularly for average and low achievers. Recent advancements in GPT-based assistants may offer adaptive, real-time feedback to address these limitations. This study employed a stratified randomized controlled design with 83 undergraduates (experimental n = 42; control n = 41) enrolled in an embedded-AI VR course. Learners completed SDVR units on embedded-AI hardware and block-based programming, followed by two hands-on tasks (EAI assembly and EAI programming). Primary outcome measures included SDL abilities, learning motivation, cognitive levels, and hands-on performance. Compared with traditional feedback, GPT-based feedback was associated with higher post-test SDL, motivation, cognitive level, and hands-on scores, with the most substantial gains observed among average and low achievers (LAs). The GPT-based feedback showed significant main effects across all outcomes (p < .001), explaining a substantial portion of the variance in cognitive level ( ω _p^2 = 0.397), SDL abilities ( ω _p^2 = 0.299), and motivation ( ω _p^2 = 0.397), representing large effect sizes. These findings demonstrate that integrating GPT-based adaptive feedback into SDVR environments provides timely and personalized support that enhances SDL, motivation, and higher-level cognitive and practical performance, particularly for learners with weaker prior achievement. The study highlights the potential of generative AI to support more personalized and equitable learning in immersive VR settings.
Herein, Pristine barium oxide (BaO) and zinc-doped barium oxide nanorods (Zn–BaO NRs) were fabricated using a facile chemical co-precipitation approach. Density of states (DOS) analysis shows that Zn incorporation in BaO introduces electronic states near the Fermi level, enhancing charge carrier mobility and promoting efficient charge separation. In photocatalytic tests, Zn–BaO NRs demonstrated superior activity under UV irradiation, achieving 99.05
Multi-source interval-incomplete data are widely encountered in real-world applications, such as medical testing, climate monitoring, remote sensing, and economic analysis. However, some of these data sources may have relatively low importance, or even no practical value. Consequently, how to effectively perform information fusion and attribute reduction on multi-source data remains a critical challenge. This paper proposes an adaptive swarm intelligence attribute selection method for a multi-source incomplete interval-value data based on conditional information amount and mutual information. First, the metric formulas on single-source incomplete interval-valued data are established, and the neighborhood granularity structure with respect to an adjustable parameter is constructed accordingly to the defined metric. Then, a fusion method based on the minimizing of conditional information amount is presented to fuse a multi-source incomplete interval-value data into a single-source incomplete interval-valued data. This method is able to select important and reliable information sources. To identify the most effective subset of features, two adaptive strategies are incorporated into the standard whale optimization algorithm (WOA) to improve its parameter selection process. Without modifying the original search operators, an adaptive WOA-based attribute selection method is developed by leveraging mutual information. The proposed method focuses on improving the robustness and effectiveness of the attribute selection process. Finally, comprehensive experiments are conducted on 12 benchmark datasets to evaluate the effectiveness of the proposed method. The results show that the proposed information fusion method has certain advantages in terms of approximate classification accuracy and quality, while the designed attribute selection algorithm surpasses several state-of-the-art methods in classification accuracy, with statistical analyses further confirming its advantage.
Classifying code snippet-based questions is essential for teaching, preparing assessment materials, and supporting intelligent learning systems in programming education. Traditional frequency-based encodings, such as TF-IDF, often fail to capture the contextual semantics within code-related questions. This study employs contextualized embeddings generated by the large language model Text-Embedding-3-Large (TE3L) to evaluate their effectiveness in classifying code-related questions. It further investigates which classifier architecture best complements the TE3L representation. Using a small-scale dataset of 171 SQL certification-style questions representative of course-level repositories, we analyze the classification complexity reduced by the TE3L scheme compared to TF-IDF. Then, we investigate classification performance under various classifier architectures with TE3L embeddings, including single models, boosting, and stacking ensembles. Results demonstrate that the TE3L scheme significantly reduces classification complexity and improves performance compared to the TF-IDF. Single classifiers, particularly the support vector machine with a linear kernel and the stochastic gradient descent classifiers, performed the best with the TE3L scheme and achieved an 11-percentage-point relative improvement over the benchmark in the weighted macro-average F1 score. The boosting and stacking techniques did not enhance performance, reflecting the challenges of ensemble learning under small-sample, imbalanced conditions. This work highlights the practical value of using LLM-based embeddings to automate question classification in low-resource educational contexts, supporting teachers in building intelligent assessment tools without requiring deep expertise in NLP or machine learning.
