Abstract: In order to accurately predict the volume fraction of dissolved gas in transformer oil and overcome the limitation that only using a single prediction model will facilitate a poor prediction accuracy and generalization performance, we put forward a prediction method of dissolved gas in transformer oil based on improved complete ensemble empirical mode decomposition (ICEEMDAN) and gray relational coefficient time-varying weight integrated prediction model. Firstly, the dissolved gas volume fraction sequences were decomposed into a series of subsequences with different time scales. Then, the gated recurrent unit (GRU) and sparrow search algorithm optimized support vector machine(SVM) were used to train each subsequence, and an integrated prediction model was combined; moreover, the prediction accuracy of different prediction methods were compared to calculate the time-varying weight of gray relational coefficient and form the prediction results of each sub series. Finally, the prediction results of each subsequence were superposed and reconstructed to obtain the final prediction results. The analysis result of the example shows that the root mean square error (RMSE), average absolute error (MAE) and correlation coefficient (R²) of single step prediction of this model are 0.593, 0.422, and 0.768, respectively, which has significantly improved the prediction accuracy compared with other algorithms, and it has a strong generalization performance. Therefore, the research results can provide a basis for the internal state monitoring of oil immersed transformers.