Abstract: The uncertainty of the air conditioning loads in buildings should be paid more attention to in the power system operation and planning. The uncertain scenarios of the air- conditioning loads in buildings can be transformed into a scenario generation problem of multiple deterministic scenarios. The basic analytic framework of the building air conditioning load scenario generation is put forward in this paper. The energy consumption characteristics of the building air conditioning loads are deeply analyzed, and the dynamic and static characteristics contained in the energy consumption time series data of the building air conditioning loads are excavated. Taking the dynamic and static characteristics of the building air conditioning load data as the conditional supervision items, the joint training loss function and the global optimization loss function are designed by combining the unsupervised confrontation training with the supervised one. On this foundation, a building air conditioning load scene generation based on the conditional TimeGAN (Time Series Generative Adversarial Nets) is proposed.Finally, an example is given to verify the feasibility and effectiveness of the proposed method. This study is of positive significance for the air conditioning loads in buildings actively participating in the operation planning of the power system.