Improving Power Transformer Lifetime Prediction using Hyperparameter Optimization

Abstract

Since the global pandemic has significantly impacted human life, technology has become a vital role in various sectors. The more technology used, the more we need the electricity supply. The stability of the electricity supply is an absolute thing to customers. Power Transformer is essential equipment for delivering electricity to customers. So the condition of the power transformers should be an essential thing that must be considered. Deep Learning is part of artificial intelligence that is widely applied to facilitate the human need. Based on its role, accuracy in the prediction results will be absolute. Hyperparameter optimization is the essential methods in the machine learning process. Errors in assigning hypermeter values can harm producing predictive values. This study discusses how to optimize the prediction results of the lifetime prediction on a power transformer. With optimal prediction results, it can help electricity management companies monitor conditions. Thereby minimizing the risk of disruption of electricity supply to customers. Models are tested and verified using a real dataset from a power transformer in several locations. The best hyper-parameter for this dataset is Bayesian Search, producing 0,001343 for Mean Absolute Error (MAE).