Abstract: Adaptive facades that integrate dynamic photovoltaic shading systems and facade planting systems offer new opportunities for urban sustainable development. Through the integration of light, wind, and heat production, adaptive facades can improve indoor environmental quality and generate electricity and crops, thereby reducing buildings' reliance on external resources. However, various environmental factors influence adaptive facades' performance, leading to significant discrepancies between traditional simulation methods and actual results. Predicting electrical energy and crop output quickly and accurately in the design stage has become a key challenge. To address this issue, an output prediction method based on machine learning neural networks was proposed with comprehensive consideration of influences on the urban built environment for adaptive facade dynamic photovoltaic shading and facade planting of urban residential buildings. This method is expected to replace traditional photovoltaic software simulation and crop output estimation methods. Specifically, this method trains an artificial neural network based on measurement data to develop two prediction models. 
　　The first model (prediction model of environmental elements and dynamic photovoltaic shading electricity output elements) used Pearson correlation analysis to obtain three environmental factors and one photovoltaic shading power output factor to train and establish the artificial neural network model. Then, the shadow loss coefficient was added to establish the prediction model. Similarly, the second model (prediction model of environmental elements and facade planting crop output elements) sought to establish the correlation between built environmental factors and output factors and choose the optimal model through the comparative selection of difference activation function. An interactive interface prediction platform was established to improve the convenience of the prediction process and the reliability of results. Based on the Rhinoceros + Grasshopper tool, the platform—which is characteristic of a menu-based operation interface, interactive information transmission, and real-time evaluation feedback—improves user-friendliness for architects and promotes the application of adaptive facade design. Due to the small difference in urban climate conditions between Haikou and Singapore, the common composite high-rise residential buildings in Singapore and Haikou in recent years were selected as the architectural carrier of adaptive facade application.
　　The results demonstrate that implementing an adaptive facade can significantly enhance indoor environmental comfort, accompanied by substantial outputs. The annual power generation of monocrystal silicon photovoltaic modules and GIGS thin film photovoltaic modules of adaptive facades on residential buildings in Singapore was 253.9 kwh and 216.7 kwh, respectively. The crop output was 99 kg per year. The annual power generation of monocrystal silicon photovoltaic modules and GIGS thin film photovoltaic modules of adaptive facades on residential buildings in Haikou were 229.6 kwh and 197.4 kwh, respectively. The crop output was 85.5 kg per year. According to calculation, it is projected that the adaptive facade can fulfill approximately 9.3%~10.9% (Singapore) and 8.4%~9.8% (Haikou) of household electricity demand and meet around 32% (Singapore) and 27.6% (Haikou) of annual vegetable demand for households. 
　　This method demonstrates the convenience of the prediction process and the reliability of prediction results. It confirms the crucial role of adaptive facades in reducing energy demands for residential buildings, enhancing urban food security, and improving indoor visual and thermal comfort. Further, it provides powerful support in designing and applying adaptive facades. With technological progress and application promotion, adaptive facade will become more and more important in future urban construction, and is expected to attract more researchers and practitioners to promote the sustainable development of cities.

Key words: adaptive facade, urban residential building, neural network, building-photovoltaic integration, building-agriculture integration, prediction method

摘要： 作为应对环境与能源问题的解决办法，耦合动态光伏遮阳与建筑表皮种植的建筑自适应表皮（Adaptive Facade）为城市可持续性提供了新机会，但如何快速准确预测电能与作物产出是设计前期关键问题之一。为解决此问题，以城市居住建筑为例，提出基于机器学习神经网络模型的产出性能预测方法，以替代传统光伏软件模拟与作物产出估算方法。首先建立由实测数据训练并进行差异性激活函数对比择优的机器学习神经网络预测模型，进而搭建交互界面预测平台。结果显示，与基础案例相比，建筑自适应表皮显著提高室内热舒适时间比，降低室内眩光，且满足家庭年用电需求9.3%~10.9%（新加坡）、8.4%~9.8%（海口）以及家庭全年蔬菜需求32%（新加坡）、27.6%（海口），该预测方法展现了预测过程的便捷性与预测结果的可靠性，推动了建筑自适应表皮在可持续城市人居环境建设领域的应用。

关键词: 建筑自适应表皮, 城市居住建筑, 神经网络, 建筑光伏一体化, 建筑农业一体化, 预测方法