Prediction of dissolved gas in transformer oil based on SSA-LSTM model

In order to accurately predict dissolved gas concentration in transformer oil and to anticipate transformer faults, a prediction model based on historical dissolved gas data in transformer oil is proposed by combining Sparrow Search Algorithm (SSA) and Long Short-Term Memory network (LSTM). First, the concentrations of dissolved gases (H
2
, CH
4
, C
2
H
6
, C
2
H
4
and H
2
) in oil were monitored and collected continuously for one year from the same transformer as the training and testing dataset of SSA-LSTM model. Then, SSA algorithm was employed to optimize the parameters of LSTM, including the number of hidden units, the maximum training cycle, the initial learning rate and so forth. By comparing and analyzing different ratios of the training and testing sets, 80% of the entire dataset was chosen as the best training set due to its good balance between prediction accuracy and convergence time. Finally, SSA-LSTM was used to predict the gas concentration in the oil over the next 7 days. The results reveal that, in comparison with traditional prediction methods (BP and LSTM), our proposed SSA-LSTM model has a better prediction performance as measured by prediction accuracy (Acc), mean absolute error (MAE), root mean square error (RMSE) and determination coefficient (R
²
). In conclusion, our proposed model can more accurately describe the variation rule of the dissolved gas concentration in the oil and provide a strong guarantee for the safe and stable operation of the power transformer.