Prediction Of Dissolved Gases In Transformer Oil Using Machine Learning For Estimating Heath Of Transformer

A transformer is the hub that connects power plants to the electrical grid.Its operational status is crucial for the stability and safety of the entire power system,making it essential for the reliability and stability of the electrical system.Dissolved gas analysis(DGA)in transformer oil is an effective method for detecting early faults in transformers,and it has been widely used both domestically and internationally.Abnormal changes in the concentration of dissolved gases in the oil can provide early indications of the transformer’s operating condition,often indicating the occurrence of internal faults such as discharge and overheating.Timely intervention can prevent more significant accidents.Therefore,DGA plays a vital role in guiding transformer maintenance work and preventing maj or power grid accidents.This study aims to use different machine learning models,including the sliding average baseline model,RNN model,improved RNN model,1D CNN+LSTM model,and improved 1D CNN+LSTM model,to predict the concentrations of ten different gases(ethylene,methane,acetylene,hydrogen,ethane,total hydrocarbons,carbon dioxide,carbon monoxide,oxygen,and nitrogen)at three different voltage levels(1000 kV,500 kV,and 220 kV)to detect early conditions that may lead to transformer failures.The results indicate that the voltage level is a key factor in determining the most effective predictive machine learning model.Overall,the R-squared value of the RNN model is better than the slidin g average baseline model and even outperforms the improved RNN model.Under the conditions of 1000 kV and 500 kV,the improved 1D CNN+LSTM model has a smaller mean absolute error(MAE)value,while under the condition of 220 kV,the improved 1D CNN+LSTM model has a larger MAE value than the 1D CNN+LSTM model.