原文信息：
Assessing building-integrated photovoltaic potential in dense urban areas using a multi-channel single-dimensional convolutional neural network model
原文链接：
https://www.sciencedirect.com/science/article/pii/S0306261924020993
Highlights
• 该研究提出了一种基于城市点云数据和卷积神经网络（CNN）的回归预测模型，用于评估中大尺度街区的建筑光伏一体化（BIPV）潜力。
• 该研究结合高斯混合模型、点云表面采样与法线估计，以及地理朝向计算，增强了卷积神经网络（CNNs）对城市空间和个体物理特征的学习能力。
• 通过结合城市建筑空间与个体特征，提高了模型对实体及朝向等物理信息的捕捉，从而提高了街区光伏潜力预测的准确性。
• 该模型框为决策者在城市环境中分阶段推广BIPV以及制定高效的能源部署策略提供了有力支持。
Abstract
Assessing BIPV (Building Integrated Photovoltaic) potential is of great significance for the comprehensive promotion and deployment of solar energy. Traditional models mostly rely on morphological parameters for PV potential assessment, presenting challenges such as subjective knowledge of urban forms and difficulty in generalization within dense urban areas. This study employs Convolutional Neural Network (CNN) for 3D modeling to evaluate BIPV potential at medium and large urban scales, introducing a framework for a multi-dimensional single-channel one-dimensional CNN model. The model utilizes the Gaussian Mixture Model combined with building point cloud data to extract the building window-to-wall ratio, thereby enhancing individual features in the building cluster point cloud. It also utilizes the 3D physical model to extract building geographic orientation information, integrating point cloud distribution through spatial connectivity to address the issue of missing geographic orientation due to rotational invariance of point cloud convolution. Additionally, it uses the surface area of the 3D model as the weight for surface point cloud sampling and combines it with normal estimation to retain building entity information, solving the disorder of point cloud convolution. This modeling framework enables accurate prediction of PV potentials in urban blocks by utilizing city point cloud data and predicting urban block boundaries. Using Melbourne City as a case study, the model demonstrates superior performance compared to traditional morphological parameter-based prediction models, with a root mean square error of 2415.548 kWh/year and an R2 SCORE of 0.937 in 75 training sets. The proposed modeling framework enables the prediction of multi-scale BIPV potential, which is beneficial for the staged promotion of BIPV and the development of effective energy deployment strategies. This study offers new insights for urban building energy modeling, deep learning, and energy prediction in complex scenarios at medium and large scales for sustainable urban development.
Keywords
BIPV
Deep learning
Urban 3D model
Window-To-Wall ratio
Building orientation
Graphics
Graphical abstract
Fig. 3. Deep learning model framework.
Fig. 5. Analysis of total PV potential of urban blocks and the influence of urban
Fig. 6. Model evaluation and comparison
团队介绍
本研究由武汉大学的研究人员共同完成。
第一作者：耿晓天，武汉大学城市设计学院建筑学博士生；通讯作者：苟中华，武汉大学城市设计学院教授，连续4年入选Top2%科学家、建筑学高被引学者，主要研究领域为建筑性模拟与优化、建筑光伏一体化设计等；其他作者包括蔡森竑，武汉大学城市设计学院建筑学博士生。
关于Applied Energy
本期小编：李辉；审核人：冯玉鹏
《Applied Energy》是世界能源领域著名学术期刊，在全球出版巨头爱思唯尔 (Elsevier) 旗下，1975年创刊，影响因子10.1，CiteScore 21.2，本刊旨在为清洁能源转换技术、能源过程和系统优化、能源效率、智慧能源、环境污染物及温室气体减排、能源与其他学科交叉融合、以及能源可持续发展等领域提供交流分享和合作的平台。开源（Open Access）姊妹新刊《Advances in Applied Energy》影响因子13.0，CiteScore 23.9。全部论文可以免费下载。在《Applied Energy》的成功经验基础上，致力于发表应用能源领域顶尖科研成果，并为广大科研人员提供一个快速权威的学术交流和发表平台，欢迎关注！
公众号团队小编招募长期开放，欢迎发送自我简介（含教育背景、研究方向等内容）至wechat@applied-energy.org
点击“阅读原文”
喜欢我们的内容？
点个“赞”或者“再看”支持下吧！
