Gas Concentration Prediction of Power System Based on EMA-Autoformer

As a core component in power systems, the monitoring and early warning of electric transformer operation status are of great importance. Oil chromatography analysis is an effective method for detecting internal faults of electric transformers, and accurate prediction of oil chromatography gas concentration is crucial for judging the operation status of electric transformers and detecting potential faults. This study proposes an EMA-Autoformer algorithm based on an improved model with self-attention and exponential moving average. Firstly, the exponential moving average method is used to fill the missing values in oil chromatography gas data. Secondly, the self-attention mechanism based on auto-correlation algorithm in the Autoformer model is employed to extract the time correlation of gas concentration time series, enabling accurate prediction of gas concentration. Experimental results show that the accuracy of the EMA-Autoformer algorithm in predicting oil chromatography gas concentration significantly surpasses that of time series prediction models such as Autoformer, Refomer, and Informer. Moreover, compared to Refomer and Informer, Autoformer performs better on the gas concentration data processed by EMA.