any historical load data, a gated recurrent unitIn view of the difficulty of new community without（GRU）load predictionalgorithm based on data transfer of similar community characteristicsis proposed to realize the load prediction of new community. Firstly,the idea of transfer learning is used to transfer the data informationwhich is highly similar to the features of the new community and predict the model parameters；Secondly, the feature data set is used asthe training set to complete the training of extreme gradient boosting（XGBoost）regression model；Then, GRU neural network is used tomodel the training sample set. When the prediction accuracy of themodel is reached, the medium and long term load prediction of thenew community with time series relationship is completed. Finally,taking a community in Lianyungang as an example, the load prediction results of the residential area form January 2020 to November2022 are obtained, so as to verify the validity of the established medium and long term load prediction model with time series relationship.