Abstract: In recent years, significant progresses in integrated energy system (IES) optimization scheduling based on deep reinforcement learning (DRL) have been achieved. However, with the development of IES, the drawbacks of DRL such as long training time and high design complexity gradually arise. Therefore, a generative adversarial imitation learning method for IES energy optimization scheduling is proposed. First, the IES intelligence learns the action exploration process adaptively by imitating the expert strategy to avoid the waste of time and computing power. Second, a discriminator network is added to discriminate the difference between the generative and expert strategies, which is used as an internal reward function to assist the neural network parameter update and avoid the influence of subjective preference and experience limitation on the IES scheduling results. Finally, the analysis based on the electric-thermal coupled system example shows that the convergence speed of the proposed method is 52% higher than that of the traditional DRL algorithm, and the convergence effect is 10% higher, while the intelligence obtains the decision-making ability close to the expert experience. The online application can realize real-time decision-making without accurate prediction and precise modeling of the environment.