TY - JOUR T1 - Transient Signal Analysis-based Optimized ANN for Early Detection and Localization of Faults in 330kV Transformers in Nigeria AU - Umuroh, Oghoghome JO - Journal of Engineering and Applied Sciences VL - 20 IS - 1 SP - 9 EP - 17 PY - 2025 DA - 2001/08/19 SN - 1816-949x DO - makjeas.2025.9.17 UR - https://makhillpublications.co/view-article.php?doi=makjeas.2025.9.17 KW - Artificial Neural Network KW - ANN KW - transformer fault detection KW - transient signal features KW - optimization KW - genetic algorithm KW - 330 kV transmission system AB -

This study developed and validated an optimized Artificial Neural Network (ANN) model for early fault detection and localization in 330 kV power transformers across Nigeria’s transmission system using a simulation-based, machine learning approach. Fault types—including line-to-ground, line-to-line, and high impedance—were simulated with IEEE 9-bus and 39-bus test systems via MATLAB/Simulink for substations such as Egbin, Gombe, Owerri, and Jos. Diagnostic features like peak voltage, rise time, crest factor, and total harmonic distortion were extracted at 100,000 samples per second and used for ANN training. A genetic algorithm (GA) optimized the ANN, improving classification accuracy and prediction performance. The model reached nearly 100% accuracy after 5000 epochs, showing strong alignment between predicted and actual fault values. Findings showed variations in model accuracy with transformer ratings, temperature, and cooling methods, peaking at 300 MVA and 28 °C under OFAF cooling. Vector field plots introduced in the study provided novel insights into F1-score transitions, clearly mapping performance shifts—an approach not present in previous research. The study concluded that integrating GA-optimized ANN models enables highly accurate, real-time fault detection and localization across Nigeria’s power grid. It also demonstrated that transformer-specific variables and environmental conditions significantly influence diagnostic precision, which is critical for efficient grid management. Based on these findings, it was recommended that Nigerian grid operators invest in ANN-based diagnostic systems integrated with SCADA for real-time fault monitoring. Additionally, transformer upgrades should consider optimal rating–cooling–temperature combinations to maximize diagnostic efficiency, while power system engineers should incorporate machine learning optimization into predictive maintenance strategies

ER -