Against the backdrop of the escalating global energy and environmental crisis, accurate prediction of air conditioning energy consumption is crucial for formulating effective energy-saving policies, optimizing energy utilization, reducing energy pressure, and reducing carbon emissions, as air conditioning energy consumption accounts for a large proportion of the entire building energy consumption. A data-driven air conditioning energy consumption prediction method is proposed based on a probability model of air-conditioning occupant behavior. Firstly, based on the analysis of the relationship between air conditioning energy consumption and factors such as air-conditioning occupant behavior, environmental parameters, time, and building, a feature label system for building air conditioning energy efficiency analysis is further constructed, covering multiple dimensions such as air conditioning occupant behavior, environment, time, and building characteristics. Secondly, the air-conditioning occupant behavior probability（AOBP）model is introduced as a factor to reflect the realtime interaction between the building environment, air conditioning occupants, and energy systems. This model considers the effects of strategy, time, events, and external stimuli, thus providing a more comprehensive estimation of air conditioning usage. Finally, particle swarm optimization algorithm is utilized to optimize the long short term memory network（LSTM）and to predict energy consumption across various building types and air conditioning systems. The simulation experiment results show that the proposed data-driven air conditioning energy consumption prediction method has made significant progress in improving prediction performance, but the calculation time has also correspondingly increased.