*Abstract*
The growing accessibility of solar photovoltaic (PV) systems offers a promising pathway for homeowners to decarbonize their buildings. However, determining the appropriate size of a PV system and battery storage remains a complex task, influenced by household energy demand, daily activity patterns, and local solar potential. This decision becomes more complex with the increasing adoption of electric vehicles (EVs), as commute patterns and charging strategies, including bidirectional charging, significantly influence electricity demand profiles.
Conventional approaches to sizing PV and battery systems rely on detailed simulations that, while accurate, are computationally intensive and often take several minutes to hours to complete. This latency reduces interactivity and limits users' ability to explore different scenarios, such as varying EV charging policies or desired levels of energy self-sufficiency.
In this work, we introduce SolarFit, an application that delivers instant, high-accuracy sizing recommendations based on simple user-provided inputs. SolarFit leverages a neural network-based surrogate model, which generates results within milliseconds. By drastically reducing computation time, our approach enables users to efficiently evaluate a range of scenarios and identify system configurations that best match their needs.
*Bio*
Julia Gschwind is a visiting Master's student at the University of Cambridge from ETH Zurich. She is supervised by Prof. Srinivasan Keshav and her research focuses on using neural networks to predict the optimal sizing of photovoltaic systems.