A data-driven multi-objective optimisation framework for energy efficiency and thermal comfort in flexible building spaces

• Formulated a novel optimisation problem for flexible building spaces incorporating movable walls. • Proposed a data-driven optimisation framework for adaptive building space configurations. • Conducted a comparative performance evaluation of optimisation algorithms, including NSGA-II, NSGA-III, R-NSGA-II, and SMS-EMOA. • Trained surrogate models using AutoGluon-Tabular, LSTM and TFT. Buildings contribute to over 30% of global energy consumption. Optimising building performance offers a significant opportunity to reduce energy consumption, and improve occupant comfort and well-being. While data-driven prediction and multi-objective optimisation have been widely studied for generative building design and control, space utilisation remains underexplored, even though nowadays more buildings are equipped with temporary walls that can flexibly divide space. The key challenges lie in how to make the best use of building-relevant data for room-level energy prediction and how to balance energy efficiency with other factors. To fill in this research gap, this paper proposes the first data-driven multi-objective optimisation framework that allows walls to be moved to the best places for optimal energy efficiency and thermal comfort. It leverages surrogate machine learning models for room-level energy prediction and Multi-objective Evolutionary Algorithms (MOEAs) enhanced with a wall-reordering mutation strategy to balance between energy usage and thermal feelings of occupants. The framework is validated through two case studies, demonstrating that the optimised configurations can reduce energy consumption by around 10% while enhancing occupants’ thermal comfort. This paper shows the effectiveness of data-driven approaches for flexible building space usage, paving the way to the sustainability and occupant well-being goals in smart buildings.