The distribution transformer (DT) is an important piece of equipment connecting users in a distribution network, and it is very important to study the law of load changes. With the promotion of IoT technology applied in a power system, more and more DTs are monitored in the distribution network, but analyzing and modeling for many devices one by one will be inefficient. Therefore, this paper proposes a technical framework of distribution transformer load forecast for cloud edge collaboration, focusing on a cluster forecast model in the cloud. First, it performs daily load curve clustering on DTs, extracts daily load patterns, analyzes the changes in daily load patterns of DTs, and puts DTs with similar power consumption behavior into one category. Then, it integrates the same type of DT load data for training, and uses the STL-LSTMs-XGBoost forecasting model to realize short-term load cluster prediction of the DT. By using the load data of a city's DTs as an example for analysis, the experimental results verify the feasibility and effectiveness of the proposed method. This work is supported by the National Key Research and Development Program of China (No. 2020YFB2009405).