Lifespan Prediction of Transformers Under Thermal and Load Stresses Using Machine Learning

In modern power systems, efficient network planning relies on the critical role of transformers, whose longevity impacts both technical reliability and economic costs. Transformers are exposed to various environmental and operational stresses, particularly thermal and loading conditions, which accelerate the degradation of their insulation systems over time. A major factor in estimating a transformer's lifespan is the deterioration of its paper insulation, where the end of this insulation’s life defines the technical end of the transformer. This study leverages machine learning, specifically a multilayer perceptron (MLP) model, to estimate the degree of insulation degradation by predicting furfural content in transformer oil—a cost-effective alternative to direct measurement—thereby facilitating accurate technical lifespan estimation. Additionally, the study incorporates a random search algorithm to address the uncertainties in insulation degradation, refining predictions of the transformer’s end-of-life. An economic model is further developed to compare the technical and economic lifespan of transformers using the Equivalent Uniform Annual Cost (EUAC) metric. Through this approach, the optimal replacement time of transformers from an economic standpoint is identified, accounting for the balancing of technical and financial considerations. Results highlight the effectiveness of advanced machine learning methodologies in improving prediction accuracy, enhancing decision-making, and reducing costs associated with transformers.