Surrogate optimization of energy retrofits in domestic building stocks using household carbon valuations

Abstract

Modelling energy retrofit adoption in domestic urban building stocks is vital for policymakers aiming to reduce emissions. The use of surrogate models to evaluate building performance combined with optimization procedures can optimize small building stocks but are insufficient at the urban scale. Recent methods train neural networks using samples of near-optimal solutions further decreasing the computational cost of optimization. However, these models do not make definitive predictions of decision makers with given environmental preferences. To rectify this, we extend the method by assigning a carbon valuation to households to derive their optimal retrofit solutions. By including the carbon valuation when training the predictive model, we can analyze the impact of households' changing attitudes to emissions. To demonstrate this method we construct an agent-based model of Nottingham, finding that simulated government campaigns to boost environmentalism improve both the number of retrofits performed and the mean emissions reduction of each installation.