Abstract: Dissolved gas analysis in oil is an effective method for early fault diagnosis of the transformer. Accurate prediction of dissolved gas volume fraction in oil can provide a theoretical basis for early fault monitoring and early warning of the transformer. In this paper, a prediction method of dissolved gases in transformer oil based on variational mode decomposition (VMD) and recurrent neural network with gated recurrent unit (GRU-RNN) was proposed. Firstly, the original time series of dissolved gases volume fraction in oil of transformer were decomposed into sub-series by VMD to reduce the influence of instability. Then, the GRU-RNN prediction method was used to predict the sub-series in one step and multi-step, respectively. Finally, the predicted sub-series were superposed and reconstructed to obtain the one-step and multi-step prediction of dissolved gas volume fraction in transformer oil. The results show that the mean absolute error (MAE) and root mean square error (RMSE) of the method are 0.057 6 and 0.068 4, respectively, and the MAE and RMSE of the multi-step prediction are 0.167 9 and 0.204 1, respectively. Compared with other prediction methods, the proposed method in this paper has a great improvement in single-step and multi-step prediction ability, which provides a reference for the power transformer monitoring and early warning.