The existing energy consumption prediction methods for office buildings fail to take into account the chaotic change characteristics of energy consumption data. In this paper, a method of energy consumption prediction for office buildings based on chaotic time series was proposed. The method first reconstructed the phase space of the time series of the research object, and judged that whether it had chaotic characteristics. Then the combination model of chaos theory was established and applied in vector regression for training. Finally, Markov chain was used to eliminate the cumulative errors caused by parameter transfer of the combination model, and the final prediction result was obtained. In order to verify the effectiveness of the algorithm, the energy consumption monitoring data of an office building in Xi'an was taken as an example for analysis. The proposed method was compared with other prediction methods, such as nonlinear autoregressive neural network and support vector regression. The experimental results show that the prediction accuracy of the chaotic time series combination model modified by Markov was significantly improved, the prediction result was better than those of other models and more consistent with the change law of energy consumption of office buildings, providing effective data support for energy conservation optimization.