Optimal design of building integrated energy systems by combining two-phase optimization and a data-driven model

By combining renewable energy systems (RESs) with buildings, building integrated energy systems (BIESs) reshape the building demand-supply relationship and contribute to both embodied and operational benefits. Their optimal designs, however, require consideration of various parameters and model complexity, making it challenging to solve multi-criteria problems and computationally intensive. This study aims to develop a datadriven two-phase optimization framework for exploring optimal BIES configurations. The novelties lie in 1) a co-simulation modelling and evaluation framework for BIES from both building and RES perspectives; 2) a twophase workflow to decouple the complex BIES model and solve multi-criteria decision problems and 3) an artificial neural network (ANN) surrogate model for rapid prediction. The proposed method is demonstrated using a prototype office building in a hot-summer and cold-winter area of China. The results show that RES-only optimization presents the lowest performances, necessitating the coordinated design for both building and RES sides. The simultaneous optimization method demonstrates the balanced quality for exploring both Pareto front and optimal solution, but encounters expensive computational costs (225 h) and convergence difficulty under 5000 optimization iterations. Comparatively, the proposed ANN-based two-phase optimization can achieve competitive solutions with half the computational time (90.6 h), which can be further reduced by efficient sampling methods. Overall, this study provides an optimization framework for holistically assessing and designing BIES that can approximate optimal configurations with significant computational savings.

Automated summary

By combining renewable energy systems with buildings, building integrated energy systems reshape the demand-supply relationship and provide both embodied and operational benefits. This study proposes a data-driven two-phase optimization framework that includes a co-simulation modelling and evaluation framework, a two-phase workflow, and an artificial neural network surrogate model. The results show that the proposed method achieves competitive solutions with significantly reduced computational time compared to other optimization methods.