Scientific literature about building occupants' behaviour and the related energy performance analyses document about several strategies to monitor window operation, including different sensors and data series lengths. In this framework, the primary goal of this study is to propose effective guidelines for minimum experiment durations and their reliability. A six-year-long database from a living laboratory was used as a benchmark; and a recursive strategy enabled to split it into more than 2,500 subsets, supporting two main steps. First, information theory concepts were used to calculate uncertainty and subsets' divergence were compared to the full database. Second, the subsets were used to train deep neural networks and evaluate the influence of monitoring lengths combined with different kinds of environmental data (i.e. indoor or outdoor). From the information-theoretic metrics, the results support that indoor-related variables can reduce most of the uncertainty related to window operation. Besides, subsets influenced by autumn and winter diverge the most compared to the full database. Considering the modelling approach, the results demonstrated that by including indoor-related variables, higher shares of reliably-performing models were achieved, and smaller subsets were needed. Seasonality has also played a major role along these lines. As a consequence, the conclusions supported the feasibility of nine -monthlong field studies, starting in summer or spring, when indoor and outdoor variables are monitored.(c) 2022 Elsevier B.V. All rights reserved.
Are years-long field studies about window operation efficient? a data- driven approach based on information theory and deep learning
Pigliautile, I;
2022-01-01
Abstract
Scientific literature about building occupants' behaviour and the related energy performance analyses document about several strategies to monitor window operation, including different sensors and data series lengths. In this framework, the primary goal of this study is to propose effective guidelines for minimum experiment durations and their reliability. A six-year-long database from a living laboratory was used as a benchmark; and a recursive strategy enabled to split it into more than 2,500 subsets, supporting two main steps. First, information theory concepts were used to calculate uncertainty and subsets' divergence were compared to the full database. Second, the subsets were used to train deep neural networks and evaluate the influence of monitoring lengths combined with different kinds of environmental data (i.e. indoor or outdoor). From the information-theoretic metrics, the results support that indoor-related variables can reduce most of the uncertainty related to window operation. Besides, subsets influenced by autumn and winter diverge the most compared to the full database. Considering the modelling approach, the results demonstrated that by including indoor-related variables, higher shares of reliably-performing models were achieved, and smaller subsets were needed. Seasonality has also played a major role along these lines. As a consequence, the conclusions supported the feasibility of nine -monthlong field studies, starting in summer or spring, when indoor and outdoor variables are monitored.(c) 2022 Elsevier B.V. All rights reserved.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.