Load forecasting is the basis of energy management and optimal scheduling of integrated energy system (IES), the forecasting accuracy is directly related to the overall operation performance of the system. This paper proposes a short-term load forecasting model for electric-thermal energy based on Transformer network and multi-task learning. Firstly, the basic architecture and theory of Transformer network and multi-task learning structure are introduced. Then, through the feature selection step based on random forest method, the typical factors reflecting the load characteristics and change law are extracted, and the input characteristics of multi-task learning are constructed. Then, the multi-task learning weight sharing layer is constructed based on transformer network, and finally the forecasting value of multi-energy load is output through the full connection layer. Finally, the effectiveness of the proposed method and algorithm is verified by the collected data from the actual micro-energy system. The results show that the proposed model can fully learn the characteristics of electricity-heat coupling and improve the accuracy of load forecasting.