I. Introduction
The HVAC systems, maintaining residents' thermal comfort in the commercial buildings, consume near half of total building energy in US [1], [2] and have great energy-saving potential. To reduce this part of energy consumption, it is required to stress on the dynamic control of HVAC systems to achieve system-level performance monitoring and performance management. And temperature and humidity as important indexes to evaluate the dynamic operation effect of HVAC systems, where the accurate forecasting is crucial to energy efficient management and occupants' thermal comfort. Traditional forecasting methods, which general consider the accurate prediction on temperature variables only, cannot achieve the prediction of temperature and relative humidity because of the strong coupling.