Novel machine learning paradigms-enabled methods for smart building operations in data-challenging contexts: Progress and perspectives

The increasing availability of building operational data has greatly encouraged the development of advanced data-driven technologies for smart building operations. Building operational data typically suffer from data quality problems, such as insufficient labeled and imbalanced data, making them incompatible with conventional machine learning algorithms. Recent advances in data science have provided novel machine learning paradigms to tackle such data challenges for practical applications, such as transfer learning, semi-supervised learning, and generative learning. This review aims to present the progress and perspectives on the effective utilization of novel machine learning paradigms for three major building energy management tasks, i.e., building energy predictions, fault detection and diagnosis, and control optimizations. In-depth discussions have been provided to demonstrate the pros and cons of different learning approaches in terms of data compatibility, modeling difficulties, and possible application scenarios, which could be helpful for enhancing the feasibility of data-driven technologies for smart building operations.