办公建筑逐时能耗异常数据在线插补方法

Online Interpolation Methods for Abnormal Data of Hourly Energy Consumption of Office Buildings

办公建筑能耗监管平台采集的能耗数据常出现突变异常、零值异常与缺失异常等典型异常数据,严重影响数据质量,难以保证数据挖掘效果。针对上述问题,通过对典型办公建筑能耗监管平台采集的逐时能耗历史数据进行大量调研分析,提出了一种基于模式分类的建筑逐时总能耗与分类分项能耗异常数据的在线识别与分步插补方法,以满足能耗监管平台异常数据的在线识别与插补要求。并将本方法与传统的拉格朗日、三次样条等插值方法进行了插值效果比较,探讨了各种插值方法对于随机缺失与不同比例连续缺失下的插值效果,以及不同模式分类方法下的不同插值模型对插值性能的影响。实验结果表明,针对常规的随机缺失异常,本方法比传统插值方法的插值误差减少了3倍以上;而对于连续缺失数据高于25%时,本方法的插值效果远优于传统方法。上述结果验证了该方法在办公建筑异常能耗数据领域的适用性,能够为充分发挥建筑能耗监管平台的作用提供有效技术手段。

Energy consumption data collected by the energy consumption monitoring platform of office buildings often includes lots of typical abnormal data,zero values,missing values and other abnormal data,which seriously affect the quality of data and make it difficult to ensure the data mining effect. To solve these problems,based on the substantial investigation and analysis of the historical hourly data collected by a typical energy consumption monitoring platform of office buildings,we proposed an online identification and step-by-step interpolation method for abnormal data of hourly total energy consumption and energy consumption of different classifications based on patterns in order to meet the online abnormal data recognition and interpolation requirements of the energy monitoring platform. And this method is also compared with the traditional Lagrange interpolation method and cubic spline interpolation method. The interpolation effect of various interpolation methods on random deletions and continuous deletion with different proportions is discussed. In the meantime,the influence of different interpolation models on interpolation performance by different pattern classification means is also explored. The experimental results indicate that the interpolation error of this method is 3 times lower than that of the traditional interpolation method or above,and that the interpolation effect of this method is far superior to the traditional method when the missing data is more than 25%. The above results verify the applicability of this method in the field of abnormal energy consumption data of office buildings and provide an effective technical means to fully exert the role of energy consumption monitoring platform of office buildings.