CH-24-C013 - A Comparative Investigation of Conventional and Data-Driven Surrogate Optimization Methods for Early Phase Building Energy Retrofit Analyses

Decarbonizing buildings is crucial in addressing pressing climate change issues. Buildings significantly contribute to global greenhouse gas emissions, and reducing their carbon footprint is essential to achieving sustainable and low-carbon cities. Upgrading buildings to enhance their energy efficiency represents a viable solution. However, building energy retrofits are complex processes that require a significant number of simulations to investigate the possible options, which limits comprehensive investigations. Surrogate models can be vital in addressing computational inefficiencies by emulating physics-based models and predicting building performance. Nonetheless, the performance and optimization of models are significantly influenced by feature engineering and different selection methods. Despite this, there is a lack of thorough research examining their impact on the effectiveness and accuracy of the models. This study proposes a modelling framework to develop surrogate predictive models for energy consumption, carbon emissions, and associated costs of building energy retrofit processes. The investigated feature selection methods are wrapper methods such as Backward-Stepwise Feature Selection (BSFS) and recursive feature elimination (RFE) that are used to select the optimal subset of features for developing the predictor models based on XGBoost and ANN architectures. Accurate surrogate models were combined with an NSGA-II optimization module to search for the nearly optimal scenarios. The models provided a comprehensive and computationally light performance, leading to the emulation of nearly 50,000 configurations within 2 minutes and identifying the nearly optimal scenarios. In addition, all developed surrogate models significantly reduced computational time compared to physics-based models, possibly identifying a set of nearly optimal scenarios. The study’s findings pave the way towards low-computational accurate models that can comprehensively predict building performance in near real-time, ultimately leading to identifying decarbonization measures at scale.