建筑供暖耗热量缺失数据改进处理方法研究

An Improved Processing Method for Replenishing Building Heating Energy Consumption Data

建筑供暖耗热量是建筑能耗的重要组成部分之一。在研究工作中,如何有效处理好建筑耗热量数据缺失情况是研究人员需解决的重要问题。针对常规回归分析进行改进,基于偏相关分析、多元线性回归分析和主成分分析构建了缺失数据处理模型,提出了建筑耗热量缺失数据的处理方法,采用2栋建筑的实测数据对提出方法的有效性进行验证。通过构建对照组模型并采用绝对相对误差指标进行分析,结果表明根据以上方法建立的改进模型可有效完成缺失数据的填补工作,具备较高的精度。

The building sector,which consists of civil buildings,together with the industry and transportation sectors,is a major consumer of energy. Building heating energy consumption is a significant component of building energy consumption in China. During researching work,energy consumption data usually miss because of some reasons,such as measuring instruments stoppage,outlier elimination,and so on. How to efficiently replenish the data set of building heating energy consumption is a key question faced by researchers. Based on traditional regression methods,an improved data pre-processing method of replenishing building heating energy consumption,which is composed by multivariable linear regression,partial regression analysis and principal component analysis,is proposed in this paper. Multivariable linear regression is the basic method used to build the replenishing model. Correlation analysis and partial correlation analysis are used to test the correlation among preliminarily selected factors and confirm the multicollinearity among different inputs. Principal component analysis is used to avoid the multicollinearity and improve the accuracy. Common outdoor inputs,such as daily mean dry-bulb temperature,daily minimum dry-bulb temperature,daily mean dew-point temperature,daily minimum dew-point temperature and daily total intensity of solar radiation,are used as potential inputs before scientific analysis. Based on simple correlation analysis,daily mean dry-bulb temperature,daily minimum dry-bulb temperature and daily total intensity of solar radiation are significantly correlated to heating energy consumption. At the same time,a new input is obtained by principal component analysis between daily mean dry-bulb temperature and daily minimum dry-bulb temperature to avoid the multicollinearity. The site measured heating energy consumption of two buildings in Tianjin is used to validate the model built by the improved data pre-processing method. Two compared models are built by traditional regression method to validate the method proposed. One model is built by simple regression analysis with outdoor daily mean dry-bulb temperature as input. The other model is built by multivariable linear regression with all five inputs without filtering and transforming. Compared with two models built by traditional regression method,the accuracy of improved method proposed in this paper is validated by the index of Absolute Relative Error. According to the comparing results,the improved model proposed in this paper can efficiently replenish the missing building heating energy consumption data with high accuracy. According to the box-whisker plot charts,the mean absolute relative indexes of improved models are about 10 % and the up whiskers of improved models are lower than compared models,which validate the high accuracy of improved method proposed in this paper. According to the analysis results,some conclusions can be obtained:(1) Finding more influencing inputs can improve the model accuracy used for replenishing building heating energy consumption. The inputs of regression models should be chosen by the correlation method.(2) Partial correlation method should be used to avoid the multicollinearity between different inputs. Principal component analysis is an applicable measure to solve this problem. This process can improve the accuracy of final replenishing model.(3) Based on two cases,the mean absolute relative indexes of models built by the method proposed in this paper is about 10 %,which can basically meet the requirements in practice.