Abstract: Accurate monitoring of sub-loads of buildings can provide information to support demand response and help improve the energy efficiency of buildings. Non-intrusive load monitoring (NILM) can obtain sub-load data from total load data, but there are relatively few studies on NILM for public buildings, and it only focuses on the disaggregation of a single type of energy. Given this, a deep learning-based electricity and gas load disaggregation method for public buildings is proposed. Firstly, the Spearman correlation coefficient is used to quantitatively evaluate the correlation between the electricity and gas sub-loads of buildings and the influencing factors, based on which feature screening is carried out. Then, the load disaggregation model is established based on convolutional neural network and bidirectional gated recurrent unit, and the attention mechanism is used to optimize the weight of the model input features to improve the accuracy of load disaggregation. Furthermore, the transfer learning technique is applied to solve the problem of insufficient training data in some buildings. Finally, the algorithm's validity is verified by real building load data and compared with related research, and the results show that the proposed method has better performance.