Paper Conference

Proceedings of BSO Conference 2022: 6th Conference of IBPSA-England

     

A Modular Thermal Space Coupling Approach for Indoor Temperature Forecasting Using Artificial Neural Networks

Jakob Bjørnskov, Muhyiddine Jradi
University of Southern Denmark

Abstract: With the increasing digitalization of buildings and the adoption of comprehensive sensing and meter- ing networks, the concept of building digital twins is emerging as a key component in future smart and energy-efficient buildings. Such digital twins enable the use of flexible and adaptable data-driven models to provide services such as automated performance monitoring and model-based operational planning in buildings. In this context, accurate indoor temper- ature models are vital to ensure that the proposed operational strategies are effective, feasible, and do not compromise indoor comfort. In this work, the significance of thermal space coupling for data-driven indoor temperature forecasting is investigated by as- sessing and comparing the performance of an isolated and coupled Long Short-Term Memory model archi- tecture across 70 spaces in a case study building. To construct the coupled architecture, an open-source tool is developed and presented, which allows the au- tomated extraction of space topology from IFC-files to identify adjacent spaces. The coupled architec- ture is found to outperform the isolated architecture for ∼84% of the investigated spaces, with significant improvements under certain operational and climatic conditions. To account for the subset of spaces where the isolated architecture performs better, it is pro- posed to select between the two architectures accord- ingly. The demonstrated modularity and embedded adaptability of the proposed model architectures pro- vide a sound basis for implementation in a highly dy- namic building Digital Twin environment.
Keywords: Artificial Neural Networks, Indoor Temperature Forecasting, Building Simulation, Data-driven modeling, Black-box modeling
Paper:
bso2022_59