Model predictive control under weather forecast uncertainty for HVAC systems in university buildings

In buildings, there are two yet conflicting optimization goals: 1) minimize energy use and energy cost and 2) maximize thermal comfort. Model predictive control (MPC) is an ideal control strategy to deal with the above conflicting optimization goals. However, one challenge hindering the implementation of the MPC in buildings is the weather forecast uncertainty. This study aimed to improve the performance of the MPC under weather forecast uncertainty by introducing an error model. The error model used a straightforward approach based on easily measurable and accessible data to improve the quality of weather forecast data. The proposed method was tested by simulation on a university building located in Norway, while the detailed information and measured data from this real building were used to develop and validate the building model used in this study. Results showed that the MPC with the error model was able to achieve almost the full theoretical potential of the MPC in terms of the energy cost and thermal comfort, with 3.4% of weekly energy cost saving and 73% of indoor temperature violation numbers reduction compared to a conventional rule-based controller. In contrast, due to the existence of weather forecast error and a lack of error addressing mechanism, the MPC without error model did not perform well and gave the energy cost saving of only 0.7% and the indoor temperature violation numbers even increased by 20%. Meanwhile, the results indicated the introduction of the error model always benefited the MPC performance even under the condition of the low error of weather forecast. This study may facilitate the real application of the MPC in buildings.