Abstract: In order to accurately predict the dissolved gas content for the transformer oil in a period of time in the future with the limited historical data, the variational mode decomposition（VMD） and the optimized support vector machine（SVM） were introduced to the prediction model. First, the original sequence was detrended by the grey model（GM）. Then, the detrended sequence was decomposed into a set of smooth modal components with the VMD. And after predicting the modal components respectively with the SVMs which were optimized by the improved non-dominated sorting genetic algorithm II, the final prediction results were obtained by reconstruction. The results show that the proposed method in this paper not only has high accuracy in gas prediction, but also can reflects the trend of gas content, and can provide a new way for the prediction in other fields of power system.