A Systematic Review of Building Energy Management and Optimization Using the Artificial Intelligence of Things (AIoT)

The transition toward a net-zero economy requires buildings to evolve from passive consumers into Grid-Interactive Efficient Buildings (GEBs). Traditional Building Energy Management Systems (BEMSs) lack the dynamic intelligence needed to control stochastic energy flows and solve multi-objective optimization problems. To systematically map this technological shift, this study conducts a Systematic Literature Review (SLR) following PRISMA guidelines, analyzing a curated corpus of 144 studies (135 primary technical papers and 9 review articles). Due to the significant diversity in methodological approaches within cyber-physical testbeds and IoT architectures discovered through the literature review process, a qualitative narrative and architectural synthesis was conducted rather than a quantitative meta-analysis. Based on this framework, this review examines emerging paradigms for Cognitive Buildings based on Artificial Intelligence of Things (AIoT), edge computing, and semantic interoperability. This review discusses the evolution of algorithms from predictive Deep Learning (DL) and Deep Reinforcement Learning (DRL) to newer approaches such as Agentic AI and Physics-Informed Neural Networks (PINNs). These new methods address the fundamental “sim-to-real” gap while ensuring thermodynamic consistency and safety in physical actuation. It also presents strategic applications in multi-objective optimization of HVAC systems, demand response, energy arbitrage, and predictive maintenance. Moreover, this review tackles major real-world deployment issues by introducing Federated Learning for data privacy, Transfer Learning for portfolio scaling, and TinyML for overcoming the computational carbon paradox of “Green AI.” By quantifying this paradox, the review contrasts the massive computational carbon footprint of cloud-based model training against the milliwatt-class efficiency of localized edge deployments. Overall, this review outlines potential research directions toward the development of autonomous Cognitive Digital Twins (CDTs) and Human-Centric Personal Comfort Models (PCMs).