LAUSR

This Topical Section collects research and application papers contributed from the International Energy Agency’s (IEA)Energy in Buildings and Communities (EBC) Annex 81 -Data-Driven Smart Buildings, on the topic of Artificial Intelligence (AI) in smart buildings. Annex 81 is a collaboration of 19 countries and over 50 organizations looking to accelerate energy efficiency adoption and demand flexibility solutions through digitalization. Annex 81 aims to investigate the potential of Software-as-a-Service innovation and intelligent data-driven building automation to reduce energy use in buildings and enable buildings to participate as distributed energy resources in support of increased use of variable renewable electricity sources. Annex 81 foresees a world where building-services IT infrastructure makes data accessible and discoverable, supporting advanced analytics, control automation, and AI-driven energy optimization. Annex 81 includes subtasks investigating fault detection and diagnosis, model predictive control, and use of buildings as distributed energy resources in electricity grids. The call for papers was distributed to Annex 81 participants. The selected papers in this Topical Section cover a broad range of AI applications in smart buildings, reflecting the broad scope of Annex 81, and provide a venue for research findings from a variety of disciplinary perspectives. Renewable energy sources account for a large proportion of future power grids. The intermittency of renewable energy generation is a major challenge, which requires comprehensive management of power generation, storage, and consumption. Understanding the key features and driving factors of building electricity use profiles is pivotal for integrated simulation with energy systems and smart grids. Zhou et al. introduce a two-tier stochastic model for typical building electricity load profile analysis and simulation. The proposed model provides an efficient tool for the optimal design of integrated energy networks and can be applied in future smart grid infrastructures. Building automatic fault detection and diagnosis (AFDD) technologies have shown great potential for energy savings. Two papers aimed at improving the performance of AFDD algorithms using AI are included here. To enable AFDD, a baseline depicting the normal operation mode is needed to detect whether the building operation deviates from normality. Literature-reported baseline development strategies suffer from scalability and cost-effectiveness issues. Huang et al. developed a data-driven method for AFDD baseline construction based on information entropy. Evaluation results indicate that the fault detection strategy adopting the proposed approach has similar or better accuracy in detecting faults compared to the same fault detection strategy using the baseline construction method from the literature. Qiu et al. propose an AFDD method, with a simple workflow and fewer sensor requirements, to detect and diagnose faults of chiller water flowmeters. AI technologies have emerged as transformational in many scientific domains. In particular, reinforcement learning (RL) has shown great success in complex control problems. Lee et al. assess deep learning-based RL algorithms in building control to better understand the pros and cons compared to the advanced rule-based control methods based on ASHRAE Guideline 36. They used the large office building energy model for cooling control in multiple climate conditions. Their study showed that the deep RL-based controls outperform rule-based methods with average energy savings of between 4 and 22%. However, they also found that a control stability issue should be carefully considered in deep RL methods, which may shorten hardware lifetimes. We want to thank all peer reviewers and all the authors for their effort. We would also like to thank the Editor and editorial staff of Science and Technology for the Built Environment for their support in disseminating the results of Annex 81.