Indochina Applied Sciences

Comparative Study of Substrate Type and Post-Annealing Temperature on Thermoelectric Properties of Bi₂Te₃ Thin Films

2025-08-10T11:02:16+07:00

This study explores the effects of substrate type and post-annealing temperature on the structural and thermoelectric properties of Bi₂Te₃ thin films prepared by DC reactive magnetron sputtering. Films were deposited on both rigid (SiO₂) and flexible (polyimide, PA) substrates, followed by annealing at 50 , 150 , and 250 °C. Structural characterization via XRD and FE-SEM indicated enhanced crystallinity with increasing temperature, particularly on SiO₂, whereas films on PA exhibited minimal structural evolution due to the substrate's lower thermal stability. EDS analysis showed that annealing shifted the Bi/Te ratio closer to the stoichiometric composition, especially for SiO₂-based films. Thermoelectric measurements revealed that SiO₂-supported films consistently demonstrated lower electrical resistivity, higher Seebeck coefficient, and superior power factor compared to those on PA. The best thermoelectric performance was observed after moderate annealing on SiO₂. These results highlight the critical role of substrate selection and thermal processing in optimizing Bi₂Te₃ thin films for both rigid and flexible thermoelectric device applications.

Hybrid PV-Reheat Thermal Power System Automatic Generation Control Using PIDD Controller Based on Hippopotamus Algorithm

2025-08-15T19:10:09+07:00

This paper examines load frequency control, an essential element of the power system that ensures frequency stability and enhances reliability, particularly in contemporary power systems incorporating renewable energy sources like photovoltaic power plants (PV) in two-area configurations and reheat thermal power plants. This research examines and contrasts metaheuristic algorithms for optimizing the settings of two sets of proportional-integral-double derivative (PIDD) controllers in regulating a two-area power plant to enhance system response. The comparative analysis of the results employed Hippopotamus algorithm (HO), Particle swarm optimization (PSO), Water cycle algorithm (WCA), and Grey wolf optimizer (GWO) to evaluate performance based on the ITAE objective function, which encompasses Overshoot, Undershoot, and Settling Time. The experimental results indicate that HO provides the lowest ITAE objective function values compared to other algorithms, exhibiting exceptional responsiveness, stability, and less settling time. This suggests that HO is an appropriate technique to be used in load frequency control for a two-area PV-reheat thermal power system.

Comparison of Physical and Optical Properties of Glass Doped with Cobalt Oxide from Chemical and Sugarcane Leaf Ash

2025-05-19T12:51:12+07:00

This research investigates and compares the physical and optical properties of cobalt oxide-doped glass prepared using conventional chemical methods and glass doped with cobalt oxide using sugarcane leaf ash as a source of SiO₂ and CaO. The composition of sugarcane leaf ash was analyzed at various sintering temperatures using X-ray fluorescence (XRF) spectroscopy, revealing a high SiO₂ content, with the maximum value reaching 69 wt%. The glass composition was formulated based on the ratio (50–x)SiO₂ (with sugarcane leaf ash used as a partial substitute for SiO₂ and CaO): 25B₂O₃: 10Na₂O: 8CaO: 7SrO: xCoO, where x represents the cobalt oxide concentration (0.00, 0.01, 0.02, 0.03, 0.04, and 0.05 mol%). The results showed that both the density and refractive index increased with higher CoO concentrations, while the molar volume decreased. The optical absorption spectra in the wavelength range of 350 – 2000 nm exhibited an increasing trend with rising CoO content. The cobalt oxide-doped glass displayed a blue color, whereas the glass doped with cobalt from sugarcane leaf ash exhibited a reddish-blue hue, as confirmed by CIE Lab* color measurements.

Solar Power Generation Prediction Using LSTM Deep Learning Algorithm in Ningxia Province, China

2025-06-03T21:21:29+07:00

The rapid expansion of photovoltaic (PV) power generation faces significant challenges due to the intermittent and stochastic characteristics of solar energy, which affect grid stability and energy management. Accurate forecasting of PV power output is crucial for optimizing grid operations and supporting the transition to clean energy. This paper proposes a deep learning approach based on Long Short-Term Memory (LSTM) networks to predict PV power generation in Ningxia Province, China. The model leverages historical power and meteorological data, which undergo comprehensive preprocessing, including outlier removal, normalization, and feature correlation analysis. The experiment is based on collecting data at 15-minute intervals, totaling 35,000 samples from a 1 MW photovoltaic power station in Ningxia for the entire year of 2023. The data include seven characteristic dimensions such as irradiance, temperature, and humidity. Comparative experiments involving Support Vector Machine (SVM), Convolutional Neural Network (CNN), and LSTM demonstrate that LSTM outperforms other methods with superior accuracy and robustness, achieving a coefficient of determination (R²) of 0.9927. The results confirm LSTM's effectiveness in capturing temporal dependencies and nonlinear patterns in PV power data. This study provides valuable insights for enhancing photovoltaic grid integration and advancing intelligent power systems.

Aspect-Based Sentiment Analysis in Thai Texts: A Comparative Study of Machine Learning and Neural Network Approaches

2025-07-26T23:13:41+07:00

Efficiently classifying messages into document categories is a fundamental task in natural language processing, crucial for organizing and extracting insights from vast amounts of textual data. This paper explores the application of machine learning algorithms, particularly neural networks incorporating contextual and linguistic semantics, for the purpose of classifying texts. Unlike traditional subject-based classification, the focus here is on overall judgment, posing unique challenges. This study examines aspect-based sentiment analysis (ABSA), which depends on accurate text classification to identify entity aspects and their associated sentiments. Using Thai language review data and a list of 400K food words, the research compares several classifiers: Naive Bayes, Linear SVM, Logistic Regression, and Bag of Words (BoW) with Keras. Results show that BoW with Keras performs best, achieving 97 % accuracy after 10 training rounds, with steady improvements in accuracy and loss reduction across epochs. This paper not only presents models and methodologies applicable to Thai-language text classification but also introduces a proposed method for measuring Thai sentiment. While the study provides valuable insights, it acknowledges the necessity for considering diverse configurations and requirements, as alternative classifiers may yield comparable or superior results. The findings herein contribute to the ongoing discourse in the field and offer a foundation for further exploration and refinement of classification techniques in Thai language text processing.