Data-Driven Strategies for Optimal Performance and Maintenance: Using Machine Learning for Improved Power Transformer Management

Power Transformers (PT) are crucial components of the electricity system infrastructure, and their failure can result insignificant costs. Balancing capital investment, maintenance costs, and operational performance is crucial for effective Asset Management. X Company currently manages an extensive portfolio of 2,362 oil-immersed PT. As PT undergo a gradual ageing process due to daily operation, monitoring their condition is paramount, achievable through oil insulation condition measurements and Dissolve Gas Analysis (DGA). DGA serves to identify potential or early faults, allowing the computation of the PT Asset Health Index (AHI). This method is auspicious as DGA data exhibits accuracy, completeness, and timeliness. Additionally, assessing the paper insulation condition, reflecting the Remaining Usage of the Lifetime (RUL) of PT is highlighted. Despite the crucial role of furan testing in evaluating transformer insulation paper quality and remaining life, it needs to be conducted periodically in X's case. Hence, this paper proposes developing a reliable machine learning algorithm that accurately forecasts furan levels using DGA test outcomes. The envisioned model will enhance insights into transformer conditions, enabling more effective and timely maintenance strategies and asset prioritization. The SVR model was the best choice in predicting furan values based on DGA variables with the lowest error rate. However, future research can be improved by incorporating data from faulty power transformers or nearing the end of their residual life. In this case, acquiring additional data on regularly conducted 2-furfuraldehyde (2FAL) measurements will enrich and validate future model development. The confidence level obtained from this model guides the implementation of priority ranking in asset management. This aids in the wise allocation of resources, maximizing operational efficiency, minimizing the risk of asset failure, and ultimately, supporting X in achieving its Asset management strategy.