Application of Deep Learning to Design Renewable Energy Systems fo a Zero Energy Multifamily Building

The Zero Energy Building design, briefly, ZEB design, is a complex optimization problem. In recent years, simulation-based optimization has demonstrated to be able to support the search for optimal design, but improvements to the method that are able to reduce the high computation are needed. This paper presents a new approach to speed up the search for optimal ZEB design solutions based on deep learning. It is applied to the problem of system design optimization for an Italian multi-family building case-study. More specifically, given a set of variables related to HVAC and renewables for describing the design space, a machine learning method that learns how to predict the objective function characterizing non-renewable primary energy consumptions is proposed. Results have shown that the approach is able to successfully drive towards optimized design (energy performance improved by 47%) while maintaining a good level of accuracy of the energy performance (error smaller than 3%) with less time and the exhaustive exploration of the design space.