| Hybrid AI Model for Optimizing Score Index of Dissolved Gas Analysis in Electric Power Transformer |
| Paper ID : 1098-ICEEM2025 (R1) |
| Authors |
|
Abdelaziz G. Saleh *1, Attia Azzam1, Gamal Attiya2, Elhossiny Ibrahim3 1Electrical Engineering Department, Faculty of Engineering, Menoufia University, Shebin Elkom, Egypt 2Department of Computer Science and Engineering, Faculty of Electronic Engineering, Menoufia University, Menouf, Egypt 3Department of Computer Science and Engineering Faculty of Electronic Engineering, Menoufia University, Menouf, Egypt |
| Abstract |
| Power transformers are essential for electrical systems, and their failures may damage power networks. Dissolved Gas Analysis (DGA) is a crucial approach for identifying transformer failures, but conventional methods face difficulties due to imbalanced data. This paper presents a hybrid AI model that includes AdaBoost and Long Short-Term Memory (LSTM) networks, improved by the Score Index Method (SIM) and Synthetic Minority Over-sampling Technique (SMOTE), to enhance fault detection accuracy. The model analyzes DGA data, addresses class imbalance, and assesses performance on Arcing, Overheating (OH), and Partial Discharge (PD) faults. The hybrid LSTM-AdaBoost model demonstrates a test accuracy of 99.21% by employing SMOTE, indicating successful classification for Arcing and OH faults. Although PD fault identification is still difficult (91.67% accuracy), the hybrid model performs better robustness than AdaBoost and standalone LSTM. For accurate transformer failure diagnosis, this method shows how well SIM may be combined with AI, providing the best option for incipient fault identification and maintenance. |
| Keywords |
| DGA, Score Index Method, SMOTE, LSTM, AdaBoost. |
| Status: Accepted |