Due to energy conservation and user needs, the set temperature and air volume of HVAC often change, which can lead to operating mode change and thus increase the difficulty of fault diagnosis. In this paper, research on fault diagnosis methods for HVAC under variable operating mode is carried out. Firstly, in order to accurately simulate the variable operating mode and typical faults of the HVAC system, a dedicated building energy simulator software TRNSYS is used for modeling of HVAC in various operating modes. Secondly, considering that the traditional PCA algorithm model, once established, could not be updated online thus cannot deal with the changes in system operating modes and generally leads to a large number of false alarms, a recursive principal component analysis (RPCA) method is developed for fault diagnosis of HVAC in varying operating modes to reduce false alarm by updating key parameters including mean and variance online. Finally, the effectiveness and superiority of the proposed method are verified by the joint simulation of TRNSYS and MATLAB.