Search

Commercial buildings are becoming smart elements that can interact with the power grid, participating in grid-level operations, meeting occupant needs and economic objectives, while enhancing the resiliency of power systems. Advanced controllers, such as model predictive control and reinforcement learning, are ideal for these novel applications which require optimal performance under varying boundary conditions, which traditional rule-based controllers are incapable of. In order to implement model predictive and reinforcement learning controllers, one of the main requirements is to have a suitable building model. However, the extensive engineering time and effort required for model development, together with the inaccuracies produced when modeling a complex real building, have delayed the deployment of advanced controllers. To overcome these difficulties, the development of accurate and computationally efficient models with low complexity, extracted from building operational data, emerges as a potential solution. The extraction of simplified models, or reduced order models, from operational data has been a key area of concern for researchers during the last decades, but its application has often been limited to specific systems, buildings, or test cases. There is a need for an open-source and standardized methodology that would render reduced order models readily available to controller designers. This would, in turn, foster the development of advanced control strategies and their application to real building systems. In this paper, a model order reduction workflow is presented as part of the development of an advanced control test bed that allows the development of advanced control algorithms for building applications. Here, several reduced order models have been created using a model order reduction method called numerical subspace state space system identification. Finally, these models have been used in the implementation of model predictive control and reinforc, Objectius de Desenvolupament Sostenible::11 - Ciutats i Comunitats Sostenibles

Clustering is an unsupervised learning technique that is useful when working with a large volume of unlabeled data. Complex dynamical systems in real life often entail data streaming from a large number of sources. Although it is desirable to use all source variables to form accurate state estimates, it is often impractical due to large computational power requirements, and sufficiently robust algorithms to handle these cases are not common. We propose a hierarchical time series clustering technique based on symbolic dynamic filtering and Granger causality, which serves as a dimensionality reduction and noise-rejection tool. Our process forms a hierarchy of variables in the multivariate time series with clustering of relevant variables at each level, thus separating out noise and less relevant variables. A new distance metric based on Granger causality is proposed and used for the time series clustering, as well as validated on empirical data sets. Experimental results from occupancy detection and building temperature estimation tasks show fidelity to the empirical data sets while maintaining state-prediction accuracy with substantially reduced data dimensionality., Comment: 6 pages, 6 figures, 1 table