Predicting Leakage Current of Distribution Insulators Based Deep Learning Gated Recurrent Unit

The serious problem with distribution insulators is the leakage current resulting from severe contaminant accumulation in coastal areas. To improve the electrical safety operations in distribution system in Taiwan, the application of a deep learning machine is developed to predict the leakage current of insulators. In this project, the deep learning methodology with a large scale of gathered data is developed to predict the insulator leakage current with weather parameters. The hourly leakage currents of 15kV distribution insulators are recorded continuously at outdoor operating conditions as the target variable for more than one year. The gated recurrent unit (GRU) based deep learning machine is utilized to predict the leakage current. The performances of GRU are compared with the RNN model with different benchmarks. The resultant experiments proved that the proposed GRU method is suitable to predict the leakage current.