Prediction of Transformer Aging Loss in 150kV Waru Substation Using GRU-LSTM Method Based on Temperature and Load

Abstract

The increasing demand for electricity in Indonesia highlights transformers' crucial role in the electrical system. This study utilizes GRU (Gated Recurrent Unit) and LSTM (Long Short-Term Memory) in a deep learning framework to predict aging losses of transformer unit 5 at the 150 kV Waru Substation. The aim is to enhance grid reliability, and efficiency, and prevent disruptions like power outages. Conducted at the 150 kV Waru Substation, the research focuses on transformer loading and temperature data. Data preprocessing involves normalizing load, oil temperature, and winding temperature data. The model architecture combines GRU and LSTM to capture short-term and long-term patterns in time series data. Training employs the Adam optimizer with customized learning rates, and performance evaluation uses metrics such as MSE, RMSE, MAPE, and MAE. Results indicate the GRU-LSTM model trained with a batch size of 64 and 75 epochs achieves superior performance: MSE of 0.0000129008474202634, RMSE of 0.00359177496793207, MAPE of 0.3943965%, and MAE of 0.00000832911556912471. This model forecasts transformer 5's aging loss over the next 30 days with an average daily deterioration rate of 0.001378178 pu/day, peaking at 0.0030481415 pu.

Keywords: Aging Loss Prediction, GRU, LSTM, Transformer