Journal of Applied Science and Technology

A COMPARATIVE EVALUATION OF MODEL ORDER REDUCTION TECHNIQUES FOR FILTER CIRCUITS

Nguyen Thi Phuong Hoa — 2026-03-12

Filter circuits often yield high-order dynamical models that impose significant computational cost. Model order reduction provides compact surrogate models that preserve dominant input-output behavior while reducing complexity. This paper presents a comparative evaluation of representative Model order reduction techniques for filter circuits, including balanced truncation, stochastic balanced truncation, modal truncation, and positive real balanced truncation. A sixth-order Chebyshev filter model is used as a benchmark. All methods are applied at identical reduced orders and evaluated using H∞ error norms together with time-domain step and impulse responses. The results highlight clear differences in absolute
and relative errors, and step and impulse responses among the considered techniques.

MISSILE GUIDANCE LAW DESIGN USING A FUZZY BRAIN EMOTIONAL LEARNING CONTROLLER

Pham Duc Hung — 2026-03-03

This study introduces an innovative guidance law utilizing a Fuzzy Brain Emotional Learning Controller (FBELC) designed for high-performance target tracking and missile homing applications. The FBELC represents a computational intelligence framework simulating the neural emotional learning mechanism, specifically the interaction between the amygdala (which governs emotional reactions) and the orbitofrontal cortex (which handles cognitive regulation). This biologically inspired structure is converted into a control system consisting of two coupled neural networks: an emotional network for rapid assessment and a sensory network for processing context. Their interplay empowers the FBELC to prioritize learning during critical moments, resulting in swifter and more resilient adaptation compared to traditional intelligent controllers. The suggested control framework utilizes the FBELC as the primary controller to emulate an ideal control strategy, while an auxiliary compensation controller is implemented to mitigate approximation errors and external perturbations, thereby guaranteeing system stability and robustness. The efficacy of the proposed FBELC-based guidance system was assessed through extensive numerical simulations and compared against the established Adaptive Fuzzy Cerebellar Model Articulation Controller (AF-CMAC) . Comparative findings indicate that the FBELC performs superiorly across critical metrics. Notably, the suggested approach yields a substantial decrease in miss-distance and exhibits accelerated convergence along with enhanced control efficiency relative to the AF-CMAC guidance law detailed in [1]. These outcomes validate that the FBELC is not only viable but also provides improved capabilities, positioning it as a highly promising candidate for modern missile guidance systems that require exceptional precision and flexibility.

SAFE & DATA-EFFICIENT SAC BASED MPPT UNDER PARTIAL SHADING WITH SIM-TO-REAL VALIDATION

Le Thi Minh Tam — 2026-03-11

This paper proposes a Safe and Data-efficient Soft Actor–Critic (SAC) based MPPT approach for photovoltaic (PV) systems operating under partial shading and dynamic irradiance. The proposed controller leverages SAC’s continuous-action learning capability to achieve fast convergence and ripple-free steady-state operation while ensuring safe duty-cycle adaptation. A comprehensive simulation benchmark comparing SAC, PID, and Intelligent Perturb & Observe (P&O) was performed using a simplified PV model with dynamic irradiance and temperature profiles. Results demonstrate that SAC achieved a 97.79% average tracking efficiency, 97.58% energy efficiency, and a significantly reduced steady-state error (2.06%) and ripple (9.70%), outperforming PID (50.05%) and P&O (41.80%) controllers. The proposed method effectively bridges the gap between AI based MPPT adaptability and real-world reliability through a safe, data-efficient learning design validated in sim-to-real scenarios.

SIMULATION STUDY OF AN IMPROVED FUZZY LOGIC CONTROLLER FOR A DC–DC BUCK CONVERTER

Do Thanh Hieu — 2026-03-10

This paper presents a simulation study on an improved fuzzy logic controller for a 100-W DC–DC buck converter to enhance output voltage regulation under varying input voltage and load conditions. The proposed controller is designed by appropriately selecting fuzzy control rules, linguistic variables, and their ranges to achieve effective nonlinear control. A detailed simulation model is developed in the MATLAB/ Simulink environment, and the control performance is evaluated in comparison with a conventional proportional–integral controller under different operating scenarios. Simulation results demonstrate that the proposed fuzzy logic controller provides a faster transient response, eliminates overshoot, and reduces
output voltage oscillations under both full-load and 50% load conditions. Furthermore, a comparison with recently reported fuzzy logic–based controllers confirms the effectiveness and robustness of the proposed approach for DC–DC buck converter applications.

CALCULATION ELECTRON TRANSPORT COEFFICIENTS IN CF4-CO2-N2 MIXTURES

Vu Trong Truong — 2026-03-12

Electron transport coefficients in CF4–CO2–N2 gas mixtures were calculated using the (MCIG) Monte-Carlo swarm code. The calculations were performed over the reduced electric field range E/N = 10 –1000 Td. The reported quantities include the energy mobility, reduced mobility (μN), mean electron energy, and the electron energy distribution function (EEDF) for mixtures containing 10%, 20%, 30%, and 40% CF4 (with CO2 and N2 in the balance according to the selected mixing ratios). The resulting dataset provides input parameters for plasma modeling and plasma processing applications using CF4–CO2–N2 mixtures.

OPTIMAL LOAD FREQUENCY CONTROL OF A THREE-AREA RENEWABLE POWER SYSTEM

Vuong- Doan Diem — 2026-03-18

This paper proposes an effective load frequency control (LFC) strategy for a three-area interconnected power system including non-reheat thermal, reheat thermal, and hydro generation units. To better represent practical operating conditions, the model incorporates key nonlinearities and constraints such as governor dead band (GDB), generation rate constraint (GRC), and the integration of solar photovoltaic (PV) and wind power. A PID/DD controller is designed and optimally tuned using the Lyrebird Optimization Algorithm (LOA) to improve the system’s dynamic frequency response. The proposed LOA-PID/DD controller is evaluated under load disturbances and compared with conventional PID/DD, PSO-PID/DD, and GWO-PID/DD controllers. Simulation results show that the proposed method provides better dynamic performance, including smaller frequency deviations, reduced tie-line power oscillations, lower overshoot, and shorter settling time. These results confirm the effectiveness of LOA in tuning PID/DD parameters for complex multi-area power systems with nonlinear constraints and renewable integration.

LIGHTWEIGHT FEDERATED LEARNING FOR GATEWAY-LEVEL DDoS DETECTION ON RESOURCE-CONSTRAINED IoT DEVICES

Le Ngoc Lanh — 2026-03-16

The rapid proliferation of Internet of Things (IoT) systems has significantly expanded the attack surface of modern networks, making distributed denial-of-service (DDoS) attacks a critical threat. Conventional intrusion detection systems are typically trained in a centralized manner, requiring raw traffic data collection from distributed gateways, which raises privacy concerns and incurs substantial communication overhead. To address these limitations, this paper proposes a lightweight federated learning (FL) framework for gateway-level DDoS detection in resource-constrained IoT environments. The framework is evaluated using three lightweight models, namely LRNet-Lite, MLPNet-Lite, and TabResNet
Lite, on the CICIoT2023 dataset deployed across a real cluster of ten Raspberry Pi 4 devices under both IID and non-IID data distributions. Experimental results show that the proposed FL framework achieves macro F1-scores of 91-92% under IID conditions and 88-89% under non-IID conditions, while maintaining inference latency below 5 ms per sample on edge devices. Furthermore, the results highlight the trade-off between model complexity and deployment cost in terms of computation, communication, and real-time performance. These findings demonstrate that lightweight federated learning provides an effective and privacy-preserving solution for practical DDoS detection at IoT gateways.

A SURVEY OF TAXONOMY-AWARE RECOMMENDER SYSTEMS IN E-COMMERCE

Dang Nhat Minh — 2026-03-16

Recommender systems are essential in modern e-commerce for alleviating information overload and delivering personalized product suggestions. However, many existing methods treat products as independent entities and ignore the hierarchical taxonomy structure that naturally organizes items, limiting their ability to generalize user preferences, especially in sparse and cold-start scenarios. This paper provides a systematic survey of taxonomy-aware recommender systems in e-commerce, categorizing existing approaches into four groups: taxonomy-based feature engineering, taxonomy-aware representation learning, taxonomy aware reasoning, and taxonomy construction and refinement. For each category, we analyze representative studies in terms of methodology, strengths, and limitations, and offer practical guidance for method selection. Finally, we highlight key challenges, including taxonomy quality, scalability, and multi-source data integration, and outline promising directions for future research.

COMPARATIVE STUDY OF MACHINE LEARNING ALGORITHMS FOR IOT CYBERSECURITY

Duy-Ngoc Nguyen — 2026-03-06

The exponential growth of the Internet of Things (IoT) has expanded global digital connectivity but simultaneously increased exposure to sophisticated cyber threats such as DDoS floods, malware propagation, and web-based exploits. To strengthen IoT cybersecurity, this study conducts a comparative analysis of representative machine learning algorithms for detecting and classifying multiple attack categories under realistic, leak-free evaluation conditions. Several widely adopted supervised models—including Logistic Regression, Decision Tree, Random Forest, Support Vector Machine, k-Nearest Neighbors, and a shallow Neural Network—were trained on the CIC IoT dataset following rigorous preprocessing and anti-leak filtering. Experimental results show that all models exhibit high reliability, achieving accuracies above 91.2 % and F1-scores exceeding 93 %. The Random Forest attains the best precision (approximately 98 %) and stability, while Logistic Regression and SVM maintain competitive accuracy with lower computational overhead, making them suitable for real-time IoT edge deployment. Overall, ensemble-based models deliver superior detection capability, whereas linear learners provide efficient and scalable alternatives for modern IoT security frameworks.

PREPAID ELECTRICITY BILL PAYMENT METHOD VIA BANK TRANSFER USING QR CODE THROUGH SEPAY PAYMENT GATEWAY INTEGRATED WITH ESP32 DEVICE

Do Tuan Khanh — 2026-03-12

The collection of electricity fees in dormitories, rental houses, and mini-apartments often faces inconveniences such as manual collection, high error rates, revenue loss, and lack of transparency. This study proposes a prepaid electricity payment method via bank transfer using QR codes through the SePay payment gateway, integrated with an ESP32 device to automatically deduct electricity costs based on the residential electricity tariff issued by Vietnam Electricity (EVN) and disconnect power when the prepaid balance is exhausted. The research methodology is based on an analysis of existing studies on IoT-based smart energy metering systems using ESP32 and prepaid meter applications. The results show that landlords generate dynamic VietQR codes for each rental room, allowing tenants to scan the QR code using e-wallets or banking applications to transfer money directly to the landlord’s bank account. The bank pushes transaction data to SePay for automatic verification, which is then stored via webhook/API and integrated with the ESP32 device to update account balances, automatically deduct electricity fees according to EVN’s tiered tariff structure, and actuate a contactor to disconnect power when the balance becomes negative. The entire process is completed in less than 10 seconds, ensuring high security and transparency. In conclusion, the proposed method is particularly suitable for dormitories, rental houses,
and mini-apartments, significantly reducing manual workload, promoting cashless payments, and enabling accurate electricity consumption control in compliance with EVN standards.

DESIGN AND EVALUATION OF ENERGY-BASED SWING-UP AND LQR CONTROL STRATEGIES FOR AN INVERTED PENDULUM ON A CART

Tran Xuan Tien — 2026-03-10

The cart-inverted pendulum is a fundamental underactuated nonlinear system, extensively utilized as a benchmark for validating control algorithms. This paper proposes a comprehensive hybrid control architecture addressing both the swing-up and upright stabilization problems. The dynamic model is derived using the Euler-Lagrange formulation and locally linearized via the Jacobian matrix. Initially, a Lyapunov-based controller pumps energy into the system, driving the total mechanical energy to the upright equilibrium. Upon entering the linear basin of attraction, the system seamlessly transitions to a fixed-gain Linear Quadratic Regulator (LQR) using a normalized angle (θwrap) logic to maintain balance. Simulations validate the hybrid algorithm and demonstrate the LQR’s enhanced performance in eliminating transient overshoot and rejecting aggressive dynamic disturbances (up to ± 3 N impulses under ± 25 N actuator saturation constraints). Quantitative evaluation further confirms that the system maintains a low angular Root Mean Square Error (RMSE) of 0.48o under persistent white noise and remains stable under parametric uncertainties of up to ±10%.

AN AUTONOMOUS LANDING ON MOVING PLATFORMS USING VISION AND ADAPTIVE CONTROL IN GPS-DENIED ENVIRONMENTS

Hugo Girard — 2026-03-10

Developing autonomous landing systems for moving targets remains a challenge in unmanned aerial vehicles (UAV), especially without reliance on global positioning system. This study proposes a vision-based control framework for multirotor UAVs, integrated into a reproducible Gazebo/PX4 simulation environment. The system utilizes a downward-facing monocular camera for relative pose estimation, coupled with an offboard velocity controller. Instead of a static control law, we introduce an adaptive PID scheme featuring gain scheduling and dead-zone handling, designed to maintain stability during the transition from high altitude tracking to precise touchdown. The landing logic is governed by a hierarchical state machine that executes a multi-phase descent, culminating in an open-loop velocity lock to counteract ground-effect disturbances. Performance validation confirms the system’s adaptability, demonstrating reliable landing capabilities on ground vehicles moving at constant velocities up to 2 m/s (7.2 km/h). These results highlight the effectiveness of the proposed adaptive control strategy in dynamic, GPS-denied environments.

REINFORCEMENT LEARNING–ASSISTED PPO–PD TRAJECTORY TRACKING CONTROL FOR A 6-DOF ROBOTIC MANIPULATOR: A COMPARATIVE SIMULATION WITH FEEDFORWARD PD

Le Thi Minh Tam — 2026-03-10

This paper presents a reinforcement learning–assisted trajectory tracking controller for a 6-DOF robotic manipulator. A hybrid Proximal Policy Optimization–Proportional–Derivative (PPO–PD) scheme is designed, where a PPO agent learns a residual torque to augment a nominal feedforward PD controller. Both controllers are evaluated in simulation on joint-space sinusoidal trajectories and their corresponding planar end-effector motions. Quantitative results demonstrate that the PPO–PD controller reduces the root-mean-square (RMS) end-effector position error from 3.95 mm to 2.38 mm (a 39.7% improvement) and decreases the peak error from 13.44 mm to 8.46 mm (a 37.1% reduction) compared with the pure
PD_forward controller. The average RMS joint-position error across six joints decreases from 0.00493 rad to 0.00298 rad (39.5% lower), while the RMS control torque decreases by 6.0% without increasing the maximum torque. These results confirm that the proposed PPO–PD controller significantly improves accuracy and efficiency over the classical PD framework while maintaining stability and interpretability suitable for industrial applications.

PEO COATINGS ON ALUMINUM SUBSTRATES FOR POWER ELECTRONIC MODULES – AN INTEGRATED SOLUTION FOR ELECTRICAL INSULATION AND THERMAL MANAGEMENT

Van-Dat Ly — 2026-03-09

Power electronics modules require dielectric layers that withstand high electric fields and dissipate heat. Polymers provide high breakdown strength but low thermal conductivity, while ceramics provide high thermal conductivity but are brittle, costly, and often require adhesive layers that increase thermal resistance. This review examines plasma electrolytic oxidation coatings on aluminum as an integrated dielectric–thermal solution for power electronic modules. It aims to position PEO relative to polymers, structural ceramics, and composite coating systems, with emphasis on the trade-off among dielectric breakdown strength, thermal conductivity, and areal thermal resistance. The formation mechanism and coating architecture are summarized, followed by a comparison with conventional dielectric materials and coating technologies. Composite PEO coatings containing thermally conductive fillers are then assessed, highlighting AlN-filled coatings with reported thermal conductivity up to 3.94 W·m⁻¹·K⁻¹. Critical needs include standardized metrology and reliability validation under humid heat, thermal shock, and power cycling.

REDUCTION OF NONLINEAR DISTORTION EFFECTS IN A PSK-OFDM SYSTEM

Nguyen Van Vinh — 2026-03-17

Nonlinear distortion caused by High-Power Amplifiers significantly degrades the performance of Orthogonal Frequency Division Multiplexing systems employing amplitude–phase modulation schemes. Due to the high peak-to-average power ratio of OFDM signals, nonlinear effects introduce constellation rotation and in-band distortion, resulting in an increased bit error rate. In this paper, a pilot-based Automatic Phase Shift method for 16-PSK OFDM systems is proposed to mitigate nonlinear distortion at the receiver side. The proposed method estimates the phase rotation by comparing transmitted and received pilot symbols and compensates for this rotation automatically without requiring prior knowledge
of amplifier characteristics or system parameters.