Prediction of Building Thermal Performance and Adaptation Measures for Populations at Risk of Heat Stress within Low-Income and Informal Urban Housing

As of 2022, the United Nations estimated that 1.12 billion urban residents lived in informal settlements. Residents of informal housing can be disproportionately exposed to overheating due to limited access to cooling, inadequate infrastructure and thermally unsuitable construction materials. Reliably predicting the thermal performance of these buildings and the impact of potential adaptations is challenging due to complex urban morphology and data sparsity, and the effectiveness of adaptations can vary widely between dwellings, even within the same dense urban block. This proposal advances work by the author on a graph schema and synthetic geometry engine capable of representing complex, multi-storey urban blocks as labelled data for EnergyPlus thermal simulations. An uncertainty-aware graph neural network will be trained on these graphs to predict changes in operative temperature and overheating degree-hours and rank mixed adaptation measures at dwelling and neighbourhood scales. This work will also test whether preserving thermally relevant topology improves generalisation to unseen settlement morphologies relative to graph-agnostic XGBoost baselines. The intended result is a rapid screening method that quantifies uncertainty and returns unsupported cases to the parent simulation or further field investigation.