ESMUST: EnergyPlus-driven surrogate model for urban surface temperature prediction

Abstract

This research presents a new urban surface temperature prediction model called the EnergyPlus-driven surrogate model for urban surface temperature (ESMUST). Estimating the urban landscape and building surface temperature is the main challenge for 3D urban structures, especially given the challenge of preparing thousands of specific inputs for the simulations. This research conducted feature engineering to identify informative input features that are mostly related to urban surface temperature sensitivity, and then generated synthetic data using EnergyPlus simulations. ESMUST was built using artificial neural networks with 42 million entries of a synthetic dataset and thirteen input features. During the validation process, ESMUST achieved 0.964% bias and efficiently predicted the surface temperature of 3D urban models with fewer inputs. The main contribution of ESMUST is the development of a surrogate model to predict 3D urban surface temperature with fewer input features and less modeling complexity. This process has great potential to save building energy and improve the outdoor thermal environment in the context of climate change. Moreover, the framework developed for ESMUST is expected to be further applied in the future to evaluate nature-based solutions such as green façades for urban comfort and sustainability.