The invention provides a power grid transformer fault classification method based on XGBoost. The power grid transformer fault classification method based on the XGBoost comprises the following steps:step 1, acquiring and integrating DGA data sets of a plurality of transformers; Step 2, performing normalization processing on the data in the DGA data set obtained in the step 1, and submitting thepreprocessed data to an XGBoost for training, and a certain number of classification regression trees are constructed to carry out fitting on the residual error of previous learning, and an optimal XGBoost parameter combination is found through grid searching, so that the diagnosis accuracy of the transformer fault is improved.