Learning-Based Hierarchical Distributed HVAC Scheduling With Operational Constraints

This investigation proposes an energy management system for large multizone commercial buildings that combines distributed optimization with the adaptive learning. While the distributed optimization provides scalability and models the fresh-air infusion as ventilation constraints, the learning algorithm simultaneously captures the influences of occupancy and user interactions. The approach employs a hierarchical architecture and uses a service-oriented framework to propose a distributed optimization method for commercial buildings. In addition, it also includes operational constraints required for optimizing the building energy consumption not studied in the literature. We show that our hierarchical architecture provides much better scalability and minimal performance loss comparable to the centralized approach. We illustrate that the influences of operational constraints on chiller, duct, damper, and ventilation are important for studying the energy savings. The energy saving potential of the proposed approach is illustrated on a 10-zone building, while its scalability is shown via simulations on a 500-zone building. To study the robustness of the approach meeting cancellations or other events that influence zone thermal dynamics, the resulting energy savings are studied. The results demonstrate the advantages of the proposed algorithm in terms of scalability, energy consumption, and robustness.