基于机器学习的外窗百叶外遮阳性能预测与敏感性研究

Prediction and Sensitivity of External Shading Performance of External Window Louver Based on Machine Learning

在建筑围护结构中越来越多地使用透明围护结构,会导致高能耗和光热环境不适等问题。为了解决这一问题,遮阳作为一种减少建筑的能源消耗和改善室内环境较为有效的手段被越来越多地使用。为探究百叶外遮阳参数对遮阳性能的影响程度,采用机器学习算法预测遮阳性能,利用基于机器学习的改进敏感性分析法探讨了影响遮阳性能的2类参数(建筑和遮阳)的局部和全局敏感性,确定影响最大的参数。研究结果表明:XGBoost预测热环境、能耗指标精度最高,而随机森林算法预测光环境指标效果最好。同时发现遮阳参数是影响室内热环境和建筑能耗的最重要因素,整体权重均在0.5以上;建筑参数显著影响室内采光,其权重高达0.9左右。

The increasing use of transparent envelope in building enclosure will lead to high energy consumption,thermal discomfort and other problems.In order to solve these problems,shading is used more and more as an effective means to reduce energy consumption of buildings and improve indoor environment.In order to explore the influence degree of louver external shading parameters on shading performance,machine learning algorithm was used to predict shading performance.Improved sensitivity analysis method based on machine learning was used to discuss the local and global sensitivity of two types of parameters(building and shading)affecting shading performance,and to determine the most influential parameters.The results show that XGBoost has the highest accuracy in predicting thermal environment and energy consumption indexes,while random forest algorithm has the best effect in predicting optical environment indexes.Meanwhile,it is found that shading parameters are the most important factors affecting indoor thermal environment and building energy consumption,and the overall weight is above 0.5.Building parameters significantly affect indoor lighting,with a weight of about 0.9.