Data-driven methods and their applications to building HVAC energy consumption prediction: A review

Abstract

In the building sector, heating, ventilation, and air conditioning (HVAC) systems account for a significant share of energy demand, especially as living standards rise in developing countries and climate change increases their demands. Accurate forecasting enables improved operational efficiency, demand response participation, and fault detection. However, existing reviews rarely focus specifically on data driven HVAC load prediction, leaving a gap in understanding the state of current methods and their limitations. This review addresses this gap by analyzing 113 publications (2019–2024), covering statistical, machine learning, and deep learning methods, including recurrent, graph, and transformer architectures. The analysis examines prediction tasks, temporal granularities, dataset properties, and feature selection practices. Results show that neural network–based models dominate the field, hourly granularity is most common, and weather, time indicators, and lagged variables are widely used features. Three main challenges emerge: (i) limited access to high quality public datasets, (ii) performance degradation due to data/concept drift in deployed models, and (iii) limited prediction of uncertainty despite its relevance for efficient planning. To address these, potential directions include transfer learning from well monitored buildings, online learning for adaptive models, and probabilistic aware approaches. By focusing specifically on HVAC load forecasting, this review provides a structured overview of current methods, highlights key limitations, and outlines research avenues to improve the accuracy, robustness, and applicability in real- world building energy management. • Literature review on the topic of energy consumption forecasting in HVAC systems • Current methodologies: applications, data resources, and emerging challenges • Machine learning applications in operational planning, maintenance and flexibility • Research gaps and opportunities on HVAC system load forecasting were highlighted • Promising emerging trends were identified to facilitate contributions in this area

Topics & Concepts