基于人员活动和建筑本体特征交互效应分析的建筑能耗归因
Attribution of building energy consumption based on interaction effect analysis between occupant behavior and building ontology characteristics
摘要：
为实现精细化公共建筑能耗管理,需充分考虑人员活动因素。本文采用基于广义多因子降维法的特征选择方法,识别具有交互效应的建筑本体和人员活动特征,采用贝叶斯方法构建了公共建筑能耗预测模型,量化交互效应。研究发现:在室时间比、人员在室率与体形系数、窗墙面积比之间具有交互效应;体形系数增大,提升了建筑能耗强度对在室时间比和人员在室率的敏感性;窗墙面积比增大,降低了建筑能耗强度对在室时间比和人员在室率的敏感性;体形系数与在室时间比、体形系数与人员在室率、窗墙面积比与在室时间比在(0,1)的值域范围内为协同效应,各交互效应归因比例平均值分别为0.41、0.56、0.40;当窗墙面积比、人员在室率取值接近1时,二者交互效应为拮抗效应,其余情况下为协同效应,归因比例平均值为0.24。
Abstract：
In order to achieve refined energy consumption management of public buildings,occupant behavior factors need to be fully considered.In this paper,a feature selection method based on the generalized multifactor dimensionality reduction method is used to identify building ontology and occupant behavior characteristics with interaction effects,and the Bayesian method is used to build a public building energy consumption prediction model to quantify the interaction effects.The research found that there are interaction effects between both the time usage intensity (TUI) and the space usage intensity (SUI),and the shape coefficient and the area ratio of window to wall (WWR).The increase of the shape coefficient increases the sensitivity of building energy consumption intensity to TUI and SUI.The increase of WWR reduces the sensitivity of building energy consumption intensity to TUI and SUI.Within the range of (0,1),the interaction effects between the shape coefficient and TUI,the shape coefficient and SUI,WWR and TUI are synergistic effects,and the average attribution percentage of each interaction effect is 0.41,0.56,and 0.40,respectively.When WWR and SUI approach 1,the interaction effect between the two factors is an antagonistic effect.In other cases,it is a synergistic effect,and the average attribution percentage is 0.24.
Keywords:energy consumption of public building; occupant behavior; characteristic; interaction effect; attribution percentage; time usage intensity; space usage intensity; shape coefficient