Building-Integrated Photovoltaic-Thermal Systems, HVAC systems, Building Modeling, System Identification, Predictive Control
Building-integrated photovoltaic-thermal (BIPV/T) systems replace conventional building cladding with solar technology that generates electricity and heat. For example, unglazed transpired solar collectors, known as UTCs, can be integrated with open-loop photovoltaic thermal (PV/T) systems to preheat ventilation air and/or to feed hot air into an air source heat pump, thus satisfying a significant part of the building’s heating and/or hot water requirements while also generating electricity. In the present study a model for a BIPV/T system with a two-stage prototype UTC integrated with PV panels has been developed and used to create a new component in TRNSYS. An open plan office space at Purdue’s Living Lab is used as a test-bed to explore system integration approaches with building HVAC systems and thermal storage mechanisms. The BIPV/T system is coupled with the building through a thermal storage tank, which serves as the heat source, and is connected to the air-to-water heat pump, for the radiant floor heating. The building ventilation system is coupled with the air outlet of the BIPV/T system. A detailed building energy model is developed in TRNSYS, which is used to evaluate the annual performance with the results showing significant energy savings. The objective is to develop models that can be implemented within a predictive control framework for the optimal set-point trajectory of the thermal storage tank. In the MPC formulation, the cost function is the integral of the electric energy consumption over the prediction horizon (48 hrs) subject to thermal comfort and equipment constraints. The study also investigates the impacts of the uncertainty in weather forecast (solar radiation) on MPC performance robustness for the integrated solar system. In our methodology, the TRNSYS model is used as a true representation of the building to identify the parameters of a 3rd order linear time invariant state-space model. The sum of squares minimization was used to identify model parameters that minimize the root-mean-squared error (RMSE) of time series predictions for the three state variables (floor surface temperature of the room; room air temperature; building envelope interior surface temperature) between the reduced order and the TRNSYS model. Known inputs to the system include the ambient temperature, solar radiation (absorbed by the envelope or transmitted through the south-facing glazed façade), internal heat gains (occupancy schedule, equipment, mechanical ventilation and infiltration) and the tank set point temperature. A pattern search optimization algorithm has been used over the training data space to identify the parameter values. Parameter bounds were set to constrain the solution space to physically plausible values. The training and calibration data sets includes 2351 (from Jan 4 to Feb 22, 7 weeks) and 1823 (from Feb 23 to Mar 30) data points, respectively. The simplified 3rd order model shows satisfactory performance with the RMSE for the three state variables within 0.5 °C. Model predictive control relevant identification methods such as 4SID (black-box identification) are also considered and the results will be compared with those using grey-box techniques.