摘要
现行建筑气候区划参考203个台站数据和地形边界等要素,具有一定的主观性,大多地区缺少真实的气象数据,亟需对空白区域进行各区划指标的空间化研究,为建筑气候区划提供客观的数据依据。空间插值是实现气象观测站点数据由点到面的主要方法。分析了经纬度、海拔、太阳辐射等要素与日平均气温的相关性,对比了多元线性回归模型和高斯过程回归模型,提出在多元线性回归模型中加入日照百分率参数可提升仅含地理信息的模型准确度,其次在高斯回归模型中仅含地理信息的回归模型精度略高于含地理信息和日照百分率的回归模型,且高斯回归模型精度显著高于线性回归模型。此外,日平均温度和日最高温度、日最低温度之间存在显著相关关系,可分别建立一元一次线性回归模型,间接获取温度日较差。最终确定了日平均温度与经纬度、海拔的高斯过程回归模型以及日最高、最低温度和日平均温度的一元一次线性回归模型,研究结果为细化建筑气候区划提供了数据支撑。
The building climate zoning boundary is divided considering the data of 203 meteorological station, topographic and so on, which is subjective and uncertain. Due to the lack of meteorological data in most areas, it is necessary to study the spatialization method of zoning indicators, so as to provide objective data basis for the determination of building climate zoning boundary. This study proposes an empirical methodology for modelling and mapping the air temperature(daily average temperature(T;), daily maximum temperature(T;), daily minimum temperature(T;)), all of which are monthly, using geographical information systems(GIS) techniques. The method can be seen as an alternative to classical spatial interpolation techniques when spatial information is available. Firstly, we have developed a regression analysis between these air temperature variables(T;, T;, T;, daily range of temperature(T;)) as the dependent ones, and some geographical variables(altitude(ALT), latitude(LAT), longitude(LON)) and solar radiation variables(solar radiation(RAD), sunshine duration(S), sunshine percentage(S1)) as the independent ones. The correlation between various variables varies greatly in different months. There is a significant positive correlation among T;, T;and T;, and the correlation coefficients are higher than 0.97. Among the three variables of solar radiation, the correlation between S1 and T;is the highest. For the three geographical variables, LON and T;are almost irrelevant.Secondly, four independent variables(ALT, LAT, LON and S1 respectively) were identified in the regression model for T;based on the correlation analysis results and the convenience of application. For the spatialization of T;, the accuracy of four regression models are compared, including two multiple linear regression models and two Gaussian process regression models. The difference is that the independent variables of each model are different. The results show that the Gaussian process regression model has the highest accuracy, with longitude, latitude and altitude as independent variables. For this model, the root mean squared error(RMSE) ranges between 0.64 and 1.09 and coefficients of determination(R;) ranges between 0.98 and 0.99 in different months. Moreover, the mean absolute errors(MAE) are evenly distributed in different region, indicating that a unified model can be used to analyze the whole country.Finally, T;and T;can establish univariate linear regression equations related to T;, respectively. The results are all very satisfactory in the case of T;and T;, with RMSE ranges between 0.33 and 1.12, and R;ranges between 0.96 and 0.99, depending on the month;while the accuracy of T;is slightly higher than that of T;. Meanwhile, for both T;and T;, the accuracy in winter and summer is significantly higher than that in spring and autumn.Based on the above results, the Gaussian process regression model of T;, as well as the univariate linear regression model of T;and T;are determined, which provide data support for the refinement of building climate zoning.
作者
吕凯琳
鲁俊忱
刘衍
杨柳
LYU Kai-lin;LU Jun-chen;LIU Yan;YANG Liu(State Key Laboratory of Green Building in Western China,Xi'an 710055,China;School of Architecture,Xi'an University of Architecture and Technology,Xi'an 710055,China)
出处
《建筑节能（中英文）》
CAS
2021年第11期52-60,共9页
Building Energy Efficiency
基金
国家自然科学基金重点项目(51838011)
国家自然科学基金面上项目(52078407)
陕西省重点研发计划项目(2018ZDCXL-SF-03-05)。
关键词
建筑气候区划
空气温度
空间插值
高斯过程回归
线性回归
building climate zoning
air temperature
spatial interpolation
Gaussian process regression
linear regression
