智能遮阳百叶眩光预测模型控制变量有效性分析  

Analysis of the Effectiveness of Control Voriables in the Intelligent Sunshade Louver Glare Prediction Model

作　　者：骆肇阳 齐轩宁 杨阳[1] LUO Zhaoyang;QI Xuanning;YANG Yang(School of Architecture at Harbin Institute of Technology,Harbin 150000,China)

机构地区：[1]哈尔滨工业大学建筑学院,黑龙江哈尔滨150000

出　　处：《照明工程学报》2024年第1期89-100,共12页China Illuminating Engineering Journal

基　　金：哈尔滨工业大学助理教授科研启动项目“基于机器学习的办公建筑室内自然采光智能调控研究”(AUGA5630109423)。

摘　　要：基于预测模型的智能遮阳百叶调控系统具有降低建筑室内眩光性能的能力,其智能性取决于调控方法。而调控方法的设计取决于对室内采光性能相关影响信息的思考。不同影响信息是调控反馈的依据,也是智能预测模型建构的基础。本研究利用机器学习技术,对不同百叶控制变量展开眩光关联性分析,以及不同控制变量在眩光影响上的交互效应。利用XGBoost构建预测模型与展开特征选择,同时采用SHAP可解释方法进一步解析变量与眩光关联大小。研究结果表明,室内使用者所在区域与遮阳百叶之间距离、遮阳形变与眩光发生最为紧密,且互动性影响明显,同时太阳高度角也是遮阳百叶重要的响应信息。结论可为智能遮阳百叶调控系统设计和预测模型建构提供参考依据。The intelligent sunshade blind control system based on the predictive model has the capability to reduce glare performance in indoor architectural spaces,and its intelligence depends on the control method employed.The design of control methods depends on thoughtful consideration of information related to the impact on indoor lighting performance.Different influencing information serves as the basis for control feedback and is also essential for constructing intelligent prediction models.This study uses machine learning techniques to conduct a glare correlation analysis for various blind control variables,as well as to explore the interactive effects of different control variables on glare impact.XGBoost was used to conduct feature selection and build a prediction model,while SHAP was used to further analyze the correlation magnitude between variables and glare.The results show that the distance between the indoor user’s location and the sunshade louvers,the deformation of the sunshade,and glare are closely related.There is a significant interactive effect,and the solar altitude angle is also an important response information for the sunshade louvers.The conclusions can provide a reference basis for the design of intelligent sunshade louver control systems and the construction of prediction models.

关 键 词：智能遮阳 预测模型 自然采光优化 机器学习 特征选择 XGBoost模型 SHAP分析 

分 类 号：TU243.2[建筑科学—建筑设计及理论]