A systematic review of building energy performance forecasting approaches

Building energy performance forecasting (BEPF) is an active area of research with the potential to improve the efficiency of building energy management systems, support global sustainability goals, and mitigate climate change impacts. This systematic review examines three main prediction methods: model-driven, data-driven, and hybrid-driven, each with different principles, basics, advantages, disadvantages, practical applications, challenges, and limitations in addressing the complexities of building energy performance. The review focuses on key influencing factors, including building features, climatic conditions, and occupant behavior, while identifying critical research gaps in current methodologies. Through a bibliometric analysis of 95 relevant publications from 2019 to 2024, this review provides a quantitative overview of research progress and emerging trends. Findings indicate that although BEPF techniques have evolved rapidly, most studies continue to overlook the variability and complexity of occupant behavior, a factor with significantly affects forecast accuracy. To address this, we propose a modular AI-integrated forecasting framework that leverages the strengths of existing approaches, integrates real-time IoT data, and incorporate advanced artificial intelligence techniques, such as generative Artificial Intelligence, reinforcement learning, and Large Language Models (LLMs). A decision-making framework is also introduced to guide method selection based on specific building characteristics, data availability, desired accuracy, and operational goals, offering practical guidance for engineering and policy applications. Additionally, future research should extend beyond individual building dynamics to include a wider range of community-level determinants, such as policy frameworks, economic factors, and social determinants of health considerations (SDOH), aiming for a more comprehensive understanding of building energy consumption patterns. This review not only synthesizes current knowledge but also lays the foundation for future innovations in BEPF. We advocate for moving towards an AI-enhanced, adaptive forecasting model that can integrate different driven methods, capture the variability and unpredictability of occupant behavior, and improve the accuracy and reliability of energy forecasts.