A study on the water content in distribution pole transformer using random forest model

Abstract

This study proposes and validates an artificial intelligence (AI)-based method for estimating the water content in the insulating oil of distribution-level transformers. The methodology includes data augmentation using noise addition, outlier removal via Isolation Forest, and data normalization through square root transformation. A Random Forest (RF) model is developed to estimate water content based on the usage period of the transformer. Correlation analyses identified the usage period as the key variable affecting water content. The model demonstrated high estimation accuracy with an R-squared value of 0.83, closely aligning estimated values with measured data. This approach provides a practical solution for real-world applications, expanding the focus to distribution-level transformers and ensuring reliable estimations through validation with actual field data. Despite limitations due to a dataset comprising 100 samples of transformer usage and oil analysis data, the method shows promise for accurate transformer lifespan assessment and efficient asset management. Future research will enhance model performance by incorporating diverse environmental conditions and comparative analyses with other machine learning (ML) algorithms, aiming to optimize estimation reliability and safety for distribution-level transformers. Consistency in the methodology description and actual models used will be maintained to avoid discrepancies.