基于卷积神经网络和遗传算法的BIPV曲面屋顶体育馆碳减排优化方法

Abstract With advances in building-integrated photovoltaic (BIPV) technology that can be applied to curved building surfaces, the carbon reduction benefits of using photovoltaics (PVs) on the large unoccupied roof areas of gymnasiums are beginning to gain traction. While existing studies have compared the energy consumption, PV power generation, and CO 2 emissions of different types of curved gymnasium roofs, the impact factors and mechanisms by which the curved surface geometries affect the three aforementioned metrics remain unclear. In this study, the energy use intensity (EUI), solar energy generation intensity (SEGI), and carbon emission intensity (CEI) of 500 groups of gymnasiums with convex, concave, hyperbolic, and free-form roof shapes were simulated respectively. The correlation between the EUI, SEGI, and CEI of the four curved gymnasium roofs was analyzed and compared. Furthermore, the point cloud transformed from the mesh intersection of the surface is fitted to a plane, which is defined as the judgement plane in this study. The influence mechanisms of the three geometric factors, namely judgement plan’s roof slope (S), slope aspect (SA), and volatility (V), on EUI, SEGI, and CEI were revealed through multivariate non-linear regression. Finally, various roof geometries were optimized by combining convolutional neural networks (CNNs) and genetic algorithms (GAs). The optimization results demonstrate that, in comparison to the initial case model, the convex, concave, hyperbolic, and free-form gymnasium roofs have the potential to reduce the CEI by up to 2.1%, 3.3%, 3.8%, and 6.5%, respectively.