Prediksi Purifikasi Minyak Isolasi Transformator Berdasarkan Hasil Uji Analisis Gas Terlarut dan Kekuatan Dielektrik Menggunakan Metode Recurrent Neural Network (RNN)

Abstract

The insulation oil within transformers can undergo chemical and physical changes, which can be indicated by dissolved gas analysis (DGA) and dielectric strength. This research aims to develop prediction models for transformer insulation oil purification using Recurrent Neural Network (RNN) and linear regression methods. DGA and dielectric strength data from PT. Solusi Bangun Andalas are utilized to train the models. The study obtained purification time predictions for insulation oil based on linear regression methods, with a dielectric strength value of 354 days having MAE 85.724693 and MSE 9436.918999, and based on Total Dissolved Combustible Gas (TDCG) value of 343 days having MAE 87.180572 and MSE 9498.780093. Using the RNN method, based on dielectric strength data, the prediction was 351 days with MAE 0.021903 and MSE 0.23994555, while based on TDCG value it was 321 days with MAE 0.62679523 and MSE 0.48404762. The research results indicate that RNN provides more accurate predictions of purification time compared to linear regression, with lower MAE and MSE. The predictions suggest that the results of dissolved gas analysis can be used as a reference for performing transformer insulation oil purification before dielectric strength values