APPLICATION OF ENSEMBLE MACHINE LEARNING METHODS IN DISSOLVED GAS ANALYSIS

ABSTRACT

Power transformers play an essential role in the electrical grid, ensuring the reliable transmission and distribution of electricity. Timely detection and diagnosis of faults in transformers are critical for preventing costly failures and ensuring uninterrupted power supply. This project investigates the application of ensemble models in dissolved gas analysis (DGA) for power transformer fault diagnosis, aiming to enhance the accuracy and reliability of fault detection. Through a comprehensive review of literature and the utilization of real-world DGA datasets, this study explores the effectiveness of ensemble algorithms, including Random Forest, AdaBoost, Stacking and Voting, in comparison to individual based models and traditional machine learning approaches. The research demonstrates that ensemble models generally outperform other methods in terms of fault classification accuracy. The findings of this research have practical implications for the power industry. Implementing ensemble models for DGA-based fault diagnosis can significantly improve the ability to detect and respond to transformer faults promptly. Recommendations include the integration of ensemble models into transformer monitoring systems, continuous data collection and analysis, interdisciplinary collaboration, and the implementation of alerting systems. Future work in this field may involve the integration of multiple sensor data sources, exploration of deep learning ensembles, addressing imbalanced datasets, and the development of real-time predictive maintenance strategies. These advancements hold the potential to further enhance the reliability and resilience of electrical systems, contributing to a more robust and dependable electrical grid.