Key points are not available for this paper at this time.
The increasing adoption of renewable energy sources (RESs) in public power grids has led to a demand for more intelligent energy management systems (EMSs) in large-scale buildings. A common approach for controlling EMSs for buildings is Model Predictive Control (MPC). For large-scale buildings, hierarchical MPC schemes have been proposed, offering the advantage of scalability through problem decomposition into multiple layers. However, hierarchical schemes often suffer from information mismatch due to information asymmetry between layers, leading to suboptimal control performance. This issue is worsened by model errors inherent to the models underlying the MPC controllers. To address these challenges, we propose a hierarchical MPC approach, which includes data-driven error compensation. Additionally, to mitigate information mismatch, a one-iteration communication step is introduced between the hierarchical layers. The proposed approach comprises two layers: an aggregator layer that controls overall energy flows of the building, and a distributor layer that allocates thermal energy to individual temperature zones. The distributor may request additional thermal budget by providing the aggregator with an otherwise expected performance loss, which it can trade off accordingly. The approach is evaluated in a software-in-the-loop (SiL) simulation using a physics-based digital twin model of a multi-zone commercial building, showing notable improvements in overall control performance in comparison to a naive hierarchical baseline and similar performance to a monolithic baseline.
Engel et al. (Tue,) studied this question.