LSTM networks for the trend prediction of gases dissolved in power transformer insulation oil

Forecasting the accurate dissolved gases content in power transformer oil can provide important basis for transformer condition assessment. In this paper, a new forecasting model based on long short-term memory networks is proposed. The sequence data of gas dissolved in oil are trained by long short-term memory network to explore the forecast characteristic. The characteristics contain the correlations between historical status monitoring information and the gases content in the forecasting time. The development trend of gases content is automatically extracted. The case studies show that the proposed method can effectively forecast gases content dissolved in transformer oil. The model achieves greater forecasting accuracy than grey model, back propagation network model and support vector machine model. It overcomes drawbacks of low stability in traditional methods and shortcoming of considering only one characteristic gas.