Accelerated inverse urban design: A multi-objective optimization method to photovoltaic power generation potential, environmental performance and economic performance in urban blocks

Urban design solutions significantly affect the green building performance of blocks. However, the current performance-based design of urban blocks mainly relies on computer simulation, which greatly reduces the decision-making efficiency of designers in weighing up different performances. This study aims to integrate machine learning predictive models with multi-objective optimisation models to improve the decision-making efficiency of designers. This study proposes a multi-objective optimisation framework integrating machine learning algorithms at the pre-design stage to enhance the PV power generation potential, environmental performance and economic performance of office blocks. Then, this paper constructed parametric models of office blocks, taking Wuhan, China, as an example and established a multi-objective optimisation model based on Rhino & Grasshopper with coupled ensemble learning algorithms. After, this paper implements a performance- driven multi-objective automatic optimisation generation design for office blocks. The results demonstrated that the ensemble learning algorithm predicted models with goodness of fit of 0.91, 0.90 and 0.92 on the test set, respectively for installation potential (IP), carbon reduction benefit (CRB) and economic payback periods (EPP). This method can be accelerated by a factor of 120-140 compared to numerical simulation methods. Compared to the performance of the initial solution in terms of EPP, IP and CRB, the Pareto set solution reduced the EPP by 13.62 %, increased the IP by 20.48 %, and increased the CRB by 19.95 %. The multi-objective optimisation framework coupled with ensemble learning algorithms proposed in this paper provides new perspectives and methods for near-zero carbon block design on a global scale.