With the advancement of energy revolution and the proposal of two-carbon goal,the integrated energy system has been paid more and more attention by many researchers. Accurate multiple load forecasting is indispensable for efficient and correct scheduling and control of integrated energy system. Based on the above requirement,the transfer learning theory is introduced,and an improved domain adaptive neural network (DaNN) load forecasting model is proposed to unified model and forecast the cooling,heating and electrical load in the integrated energy system. Firstly,the feature pictures of multiple loads are constructed by historical data and be input into the parameter sharing layer of improved DaNN. Secondly,based on the characteristics of combined forecasting of multiple loads,the loss function of traditional neural network is improved. The maximum mean difference (MMD) index is added,and the training model is optimized. Finally,the forecast values of cooling,heating and electrical loads are output through three independent full connection layers. Through the actual case verification and comparison with the traditional model,it proves that improved DaNN model effectively improves the accuracy of multiple energy load forecasting of integrated energy system.