我的訂單 | 我的收藏 | 加入收藏 | 返回首頁
暖通空調>期刊目次>2022年>第8期

基于數據挖掘的公共建筑能耗監管平臺異常數據修復研究

Research on abnormal data repair of public building energy consumption monitoring platform based on data mining

張城瑀[1] 趙天怡[1] 特日格樂[1] 馬良棟[1] 婁蘭蘭[2] 朱 凱[3]
[1]大連理工大學 [2]大連理工大學人工智能大連研究院 [3]大連群智科技有限公司

摘要:

公共建筑用能設備多、建筑面積大、使用人數多,具有較大的節能潛力。但由于建設費用有限導致的數據分項計量異常及傳感器或采集器故障導致的數據缺失和突變等問題,其配套的建筑能耗監管平臺獲取的電耗數據經常出現數據異常問題。本文研究以聚類算法為基礎,提出了一種由KNN-Matrix算法與KNN-Slope算法共同構成的異常數據修復體系。KNN-Slope算法根據異常數據當日用電趨勢線,尋找用電趨勢一致的最近歷史電耗數據,以加權計算后的電耗值作為插補值進行異常數據修復。KNN-Matrix算法引入以矩陣形式表征的用電強度量化等級,尋找量化等級與用電趨勢均一致的最近歷史數據或平均歷史數據作為插補值進行異常數據修復。結果顯示,在面向不同數據異常比例和不同公共建筑類型時,上述修復體系可使99%的異常數據在修復后與真實數據的相對誤差在30%以下,且相對誤差最大值、平均值均大幅下降。

關鍵詞:公共建筑;能耗監管;數據挖掘;臨近算法;量化等級;數據修復

Abstract:

Public buildings have many energy-using equipment, large construction areas, and a large number of users, which have great energy-saving potential. However, due to the problems of the abnormal data itemization caused by limited construction costs and the data loss and mutation caused by sensor or collector failures, the power consumption data obtained by its supporting building energy consumption monitoring platform often have anomalies. Based on the clustering algorithm, this paper proposes an abnormal data repair system composed of KNN-Matrix algorithm and KNN-Slope algorithm. Based on the current power consumption trend line of the abnormal data, the KNN-Slope algorithm looks for the recent historical power consumption data that are consistent with power consumption trend, and uses the weighted calculated power consumption value as the interpolated value to repair the abnormal data. The KNN-Matrix algorithm introduces a quantitative grade of electricity intensity characterized in matrix form, and looks for the recent historical data or average historical data that are consistent with the power consumption trend as an interpolated value for abnormal data repair. The results show that when facing different data anomalies and different public building types, the above repair system can make 99% of the abnormal data have a relative error of less than 30% with the real data after repair, and the maximum and average values of the relative errors are greatly reduced.

Keywords:publicbuilding;energyconsumptionmonitoring;datamining;proximityalgorithm;quantitativegrade;datarepair

    你還沒注冊?或者沒有登錄?這篇期刊要求至少是本站的注冊會員才能閱讀!

    如果你還沒注冊,請趕緊點此注冊吧!

    如果你已經注冊但還沒登錄,請趕緊點此登錄吧!

99香蕉国产线看观看这里有精品