Abstract: Affected by special periods such as holidays, people returning home, seasons, and distribution network planning and operation factors, distribution transformers are prone to heavy overload. However, in the actual collection process, the number of heavy overloads of distribution transformers is relatively small, resulting in the inability of the classifier to effectively identify abnormal samples. Therefore, a distribution transformer heavy overload identification method based on mixed sampling and random forest is proposed. Clustering and stratified under-sampling are performed on normal samples, and SMOTETomek sampling is performed on heavy overload samples to narrow the sample distribution gap. The sampled data set is then used to train the random forest model.