Decentralized architectures for thermal control of buildings

This paper considers the problem of partitioning a building or any other complex energy system into clusters for its decentralized thermal control. Using a Model Predictive Control (MPC) framework, a measure of deviation in performance between centralized control and decentralized control, called the Optimality Loss Factor (OLF) is derived. For a given partition size, the computationally intractable problem of determining the partition with the smallest OLF is then considered and an agglomerative clustering approach is proposed to overcome the computational limitation. The potential use of this approach to determine decentralized control architectures which yield the best trade-off between the underlying optimality and robustness objectives is demonstrated using an example.