Digital building logbook framework for energy efficiency and decarbonization using deep learning

• Develop a Digital building logbook framework to contribute to energy efficiency and decarbonization. • Incorporate deep learning to advance the capabilities of digital building logbooks. • Transfer learning facilitates digital building logbook development and adaptation. • Conduct optimization techniques to enhance the digital building logbook recommendation system. • Provide recommendations for under-design buildings and renovation procedures for existing ones to improve performance. Numerous studies have shown that energy efficiency and decarbonization are crucial for sustainable buildings. To evaluate and improve building efficiency, several tools have been developed. However, most tools only assess one building lifecycle phase. Additionally, they require significant expertise and violate occupant privacy. Hence, this paper presents a framework for generating a Digital Building Logbook (DBL) that provides five indicators to evaluate the building in terms of energy consumption and carbon emissions throughout the pre-construction and operational stages. The framework experiments with two deep-learning techniques, utilizing various configurations to produce the DBL: multi-layer perceptron and convolutional neural network models. The data used for training the models were generated using parametric simulation and lifecycle assessment, considering both internal and external environments. Furthermore, the transfer learning (TL) technique is investigated to develop new models for different building types with minimal effort. The evaluation results are used to optimize the building alternatives. Particle Swarm, Genetic Algorithm, and Ant Colony optimization techniques were evaluated to find the most suitable for the DBL. The tool issues recommendations for under-design buildings and renovation procedures for existing ones to improve performance. The presented framework is validated through a case study of a residential villa. Further, a second case of an office building is presented to validate the TL technique. The study analyzes the results to identify the most influential features of the predicted outputs using the Shapley additive explanation. The results demonstrated significant accuracy, guiding stakeholders in making decisions to attain energy efficiency and decarbonization.