Achieving precise cooling load management for grid-responsive office buildings considering occupant preference

Buildings play a critical role in demand-side management by regulating energy consumption and mitigating the imbalance between generation and load. The reinforcement learning (RL) algorithm has been widely used for demand control in building HVAC systems. However, there are few studies considering the impact of occupant override behavior in cooling load management. In this study, we considered different control schemes in three levels of detail (LoD). According to the results for two office buildings, we found that the proposed RL algorithm can significantly improve the on-track time for cooling load management (the percentage of time with cooling rate within ± 10% of the target cooling rate). For the large and the small office buildings, the RL improves the on-track time by 17% and 40%, respectively. The impact of occupant override behavior varies with thermal preference and building size. It is found that there is an obvious decrease for the “cool” group in the large office, as well as for both “median” and “cool” groups in the small office. For the “warm” group, the impact is not significant for both offices. RL can still maintain an on-track time of 92% and 80% for the large office and the small office, respectively. According to the control schedule, it is indicated that the RL is able to learn the thermal dynamics of different buildings and manage the cooling load by precisely adjusting the cooling setpoints. According to the results, it is recommended to consider occupant thermal preference for determining the targets of demand response buildings. Regarding the robustness of RL algorithms, it should be noted that although the trained RL agent shows good robustness within the tested scenarios, all results are derived solely from US data in August 2017 and that the conclusions may not directly generalize to other seasons, years, or climates. To achieve wider transferability, the RL should take more factors into account in future studies.