應用生成對抗網路建立光照需求模擬最佳之室內佈局

Abstract

室內佈局與能源光照需求有著密不可分的關係，並同時影響著建築物能源消耗。許多研究探討如何藉由改變室內布局降低能源消耗，然而，這些方法往往僅適用於初步設計階段，並無法直接且有系統地應用到實際存在的建物中，此外，也常因為使用模擬的方式在進行分析時對建物進行過多假設導致其難以反映真實情況，因此，本研究擬提出一套利用生成性對抗網路(GAN)生成室內佈局輔助光照需求分析，進而實現室內佈局的最佳化的方法，而最佳化後的結果亦能夠成為後續相關案例的參考對象。透過蒐集相關平面圖，本研究首先利用物件辨識模型提取相關有用資訊，並藉由深度學習模型生成相對應的多個室內佈局選項，再將生成的平面圖像轉換成立體的幾何模型，接著進行光照模擬分析，分析過後即可比較其與原佈局的優劣，進而給出最佳的室內佈局建議。而根據我們最後的實驗結果，改變室內佈局最多可提升38%的太陽輻射接收量，進而降低室內光照需求。

Interior building layouts and lighting energy demand are interrelated and the design of the layouts can further affect the building energy consumption. However, recent studies are limited to the preliminary design phase, which cannot be directly and clearly applied to the case of actual buildings after construction. In addition, the use of simulation methods for analysis often results in too many assumptions for buildings themselves and makes it difficult to reflect on the real situation. Therefore, this research proposes a set of indoor building layout auxiliary lighting demand analysis using a generative adversarial network (GAN) to optimize the indoor building layout. The results can also be used as a reference for subsequent related cases. This research collected hundreds of online indoor layout data as an input for the framework. This study first uses object recognition models to extract relevant space and object information from the building layouts and then generates numerous corresponding indoor building layout options by the generative model. Next, the system automatically converts the 2D floor plan into a 3D geometric model. Finally, the system imports the 3D model to the energy analysis software for lighting simulation analysis. The results can be compared with the original layout and provide optimal indoor layout suggestions. The experiments showed that the radiation reception after changing building layout can be improve by 38%.