Aiming the problem that the accuracy of current building energy consumption prediction method depends heavily on environmental parameters (air temperature, air pressure, humidity, etc.), a short-term prediction model of time series building energy consumption based on Chaos-BP is proposed. Firstly, the optimal delay time and embedding dimension of data samples are obtained by CC method. And the small data amount method is used to prove that the building energy consumption has chaotic characteristics. Secondly, the input structure of BP neural network is determined by phase space reconstruction, and again the middle school building is the research object. Finally, the Matlab software is used as the simulation tool to simulate the Chaos-BP prediction model and the BP prediction model respectively. The experimental results show that the building energy consumption has chaotic characteristics. Compared with the BP neural network prediction model, the Chaos-BP model can accurately predict the building energy consumption and provide a scientific basis for the development of building energy conservation work.