Machine assistance in energy-efficient building design: A predictive framework toward dynamic interaction with human decision-making under uncertainty

At the energy-efficient buildings design stage, architects suffer from multi-discipline requirements and insufficient information to make proper decisions during the process. Inspired by the human nervous system's estimation mechanism, we proposed a data-driven process-based framework for decision-making support. This framework achieves the performance-oriented decision aid under uncertainties based on a general component design, consisting of three parts: the probabilistic surrogate modeling, ensemble modeling, and the model interpretation method. With the characterization of uncertainties into aleatory or epistemic based on the possibility for minimization, the component’s design enables the framework to achieve dynamic interaction with users and inference toward higher intelligence to “make assumptions” in potential design space. Subsequently, it maps possible consequences of output scenarios to input variants’ causes by generating informative feedback and ensures a robust prediction under certain flexibility of incomplete inputs. We utilized the framework as an assistance system to conduct the strategic feedback of energy efficiency for building designers in different early design stages: The framework is tested on the Energy Performance Certificate (EPC) data from England and Wales (19,725,379 buildings). The result achieves a comparable forecasting performance as the SOTA machine learning and provides coherent input variants' interpretation. More importantly, during the design process, the framework enables to interactively provide building designers with expected building energy efficiency range in on-going possible design space with intervention consequences and input causes interpretation. Eventually, it drives users to operate toward higher energy-efficient building designs.