Window opening behavior significantly impacts indoor air quality, thermal comfort, and energy consumption. A field measurement was carried out in three typical rooms (a standard office, a meeting room and a smoking office) within an office building. The window state and the physical environment were continuously recorded during the measured periods. Three typical window opening behaviors were found in the measured samples, namely, active, moderate, and passive. The common logistic regression coefficient indicated that solar radiation exhibited the greatest effect on window opening behavior in the smoking office and standard office. Typically, window opening behavior in the meeting room was the most strongly correlated with time of the day, mainly because of the meeting schedule for occupants in the meeting room. This study discussed the dividing principles involved in setting the dummy variable interval level (discretizing continuous variables and dividing them into different intervals), and proposed a method to determine the optimal interval level of each variable. The improved model led to the increase in the prediction accuracy rate of the window being opened by 2.0% and 3.3% according to the comparison with the original model based on dummy variables and the common model based on continuous variables, respectively. This study can provide a reference value for simulating energy consumption in office buildings in the future.

窗户打开行为显着影响室内空气质量、热舒适度和能源消耗。在办公楼内的三个典型房间（标准办公室、会议室和吸烟室）进行了现场测量。在测量期间连续记录窗口状态和物理环境。在测量的样本中发现了三种典型的开窗行为，即主动、中等和被动。共同logistic回归系数表明，在吸烟办公室和标准办公室中，太阳辐射对开窗行为的影响最大。通常，会议室的开窗行为与一天中的时间关系最密切，这主要是因为会议室中的会议安排。本研究讨论了设置虚拟变量区间水平（将连续变量离散化并划分为不同区间）的划分原则，并提出了确定各变量最优区间水平的方法。与基于虚拟变量的原始模型和基于连续变量的普通模型相比，改进后的模型使得窗口打开的预测准确率分别提高了2.0%和3.3%。本研究可为今后办公建筑能耗模拟提供参考价值。与基于虚拟变量的原始模型和基于连续变量的普通模型相比，改进后的模型使得窗口打开的预测准确率分别提高了2.0%和3.3%。本研究可为今后办公建筑能耗模拟提供参考价值。与基于虚拟变量的原始模型和基于连续变量的普通模型相比，改进后的模型使得窗口打开的预测准确率分别提高了2.0%和3.3%。本研究可为今后办公建筑能耗模拟提供参考价值。