DEEP LEARNING APPROACHES FOR BUILDING ENERGY CONSUMPTION PREDICTION

Building energy consumption prediction has emerged as a critical component in the global effort to achieve energy efficiency and sustainability in the built environment. This systematic review comprehensively analyzes the application of deep learning approaches in building energy consumption prediction, synthesizing findings from recent research published between 2018 and 2024. Our methodology involved a systematic search across major scientific databases, including IEEE Xplore, Science Direct, and Web of Science, yielding 127 relevant studies that met our inclusion criteria. The review reveals significant advancements in prediction accuracy through various deep learning architectures, with particular success in hybrid models combining multiple neural network types. Key findings indicate that transformer-based models and attention mechanisms show superior performance for long-term predictions, while LSTM networks excel in capturing short-term consumption patterns. However, challenges persist in data quality, model interpretability, and real-world deployment. This review provides valuable insights for researchers and practitioners, highlighting promising research directions in transfer learning, explainable AI, and integration with building management systems.