Energy statistics for 2018 show that Buildings, including residential and commercial buildings, accounted for 20% of world’s total energy consumption and were responsible for 28% of energy-related CO₂ emissions globally. There is continuous progress towards sustainable buildings however it is not fast enough to outpace demand growth. Therefore, if the average global temperature rise is to be limited to less than 2°C by 2030, buildings must undergo global transformation, leading to highly energy-efficient and low-carbon buildings.

Thanks to the emergence of Internet of Things (IoT) technologies as well as the recent developments in Machine Learning (ML) and Deep Learning (DL), the concept of smart buildings and smart homes has become a reality. IoT-based energy management systems (EMS) combine advancements in sensor technologies, communications and advanced control algorithms for optimized and efficient energy use in buildings/homes. Thus, the adoption of those systems leads to economic benefits for both, the end-users and the Distribution Network Operators (DNOs) while reducing the environmental impact of the buildings sector. Over the past decade, there has been a surge in Cloud Computing and in the development of smart building/home devices that can connect to the Internet and be controlled remotely. This has led to cloud-centric IoT-based solutions for the development of smart buildings/homes. However, due to an ever-increasing number of IoT devices, network bandwidth and security weaknesses have become a bottleneck, hindering the performance of these systems. Edge computing is a recent computing paradigm that has been proposed as a potential technology for the construction of a scalable network infrastructure in the user’s vicinity, which would be rich in IoT resources (i.e. storage, compute, and bandwidth), processing real-time security-crucial data in a one-hop manner to minimize latency. Edge computing is viewed as the most promising technology for future IoT-based smart buildings. By coupling it with Distributed Ledger Technologies (DLT) such as blockchain, security and privacy may be increased further.

Recently, various ML and DL techniques have been proposed in the building sector for optimized demand response (DR), by forecasting energy consumption and generation (including Distributed Energy Resources – DER) under different operational circumstances. They operate as a black box to discover the relationship between various input features and output targets by means of a series of supervised and unsupervised learning algorithms. However, these approaches usually have two major drawbacks:

1. The energy efficiency (EE) recommendations of the resulting ML and DL models are usually only based on the collected energy data and do not take into account other important factors, such as the user’s comfort or the reputation of the recommended energy-saving measures (i.e., measures that are popular among users and that users are more likely to put into practice in the long term);
2. generally, classical ML models, especially those based on DL, cannot be interpreted, impeding the recommendation of EE measures.

On the one hand, a Social Computing approach, based on Virtual Organizations can tackle the first problem where EE measures are proposed on the basis of the preferences of users with a similar consumption behavior profile. On the other hand, recent explainable Artificial Intelligence (XAI) approaches, such as hybrid neuro-symbolic AI algorithms, increase interpretability, where Deep Symbolic Learning has emerged as a promising solution that will be investigated and integrated in the proposed Edge computing architecture.

In this 3-year project, a plug-and-play Adaptive and Intelligent Building Energy Management System (AI-BEMS) will be investigated, designed and validated by conducting two pilot tests in a real-world setting. It will be based on a secured and scalable 3-tier Edge Computing architecture supported by virtual organizations that will orchestrate distributed XAI algorithms for DR and EE optimization, integrating local generation and storage systems, enabling demand side management with dynamic and demand-dependent tariffs. The EE measures will be recommended to users by following a social computing approach, that will try to maximize indoor comfort (thermal comfort, visual comfort and indoor air quality), moreover the system will flexibly adapt to occupancy patterns, following consumers’ preferences and considering the reputation of EE actions. This novel AI-BEMS overcomes some of the drawbacks of the existing IoT systems based on cloud computing related to network bandwidth, security weaknesses and scalability of services. It implements cost-effective and high-performance prediction of demand flexibility and autonomous energy consumption optimization techniques. In addition, this AI-BEMS will have greater insight into business, offering added-value to DNOs and improved performance to the end users.