Prediction of dissolved gas content in transformer oil based on VMD-GRU

This study presents a novel prediction model founded on Variational Mode Decomposition (VMD) and Gated Recurrent Unit (GRU) techniques, aimed at addressing the challenge posed by the nonlinearity and non-stationarity of sequences on the accuracy of dissolved gas content prediction in transformer oil. Initially, the model employs the differential method to extract the trend term from the original sequence, followed by VMD decomposition of the residual sequence to derive a series of stationary modal components. Subsequently, GRU is employed to forecast each modal component, and ultimately, the prediction outcomes are reconstructed to yield the final prediction value. Experimental validation demonstrates a significant enhancement in prediction accuracy for H2 dissolved gas with the VMD-GRU model. Furthermore, simulation experiments conducted on other dissolved gas components such as CH4, CO, total hydrocarbons, etc., corroborate the efficacy of the prediction model, underscoring its broad applicability and reliability in forecasting dissolved gas content in transformer oil.