能源管理系統最常見的設備斷線遺漏值補值方式為線性內插法，有鑑於此法在斷線時間較長時會將資料數值(耗電量、及時耗功、功率因素等)呈現的波形或斜率抹平，對於後續需量計算以及契約容量設定，進而影響電費分析計算及迴歸模型建立與後續節能績效預測的結果。 本研究方法以K-最近鄰迴歸法、支援向量迴歸 、多層感知器迴歸、移動自迴歸平均方法對能源管理系統蒐集的電能資料進行補值，其目的在分析比較使用以上方法與線性內插法補值作比較，對此四種方式的結果使用MAPE進行比較分析，進而選擇該演算法進行未來系統補值改善。 本實驗結果使用K-最近鄰迴歸法以連續1日訓練資料集得到較佳的補值方法，電壓使用KNN Regression K=140，電流使用SVR cost =30，epsilon 0.3，功率因素使用SVR cost =60，epsilon 0.1，耗電量使用X-12-ARIMA，即時耗功使用SVR cost =90，epsilon 0.2得到比原來使用LI更好的補值效果。 ;The most common method for complementing the missing value of the equipment in the energy management system is linear interpolation. In view of this method, the data value (power consumption, timely power consumption, power factor, etc.) will be presented when the disconnection time is long. The waveform or slope is smoothed, which is used for subsequent demand calculation and contract capacity setting, which in turn affects the results of electricity cost analysis calculation and regression model establishment and subsequent energy conservation performance prediction. This study uses KNN Regression, SVR, MLP Regression , and ARIMA method supplements the energy data collected by the energy management system. The purpose is to compare and compare the above methods with the linear interpolation method. The results of the four methods are compared and analyzed using MAPE, and then the algorithm performs future system complement value improvement. The results of this experiment use the K-nearest neighbor regression method to obtain a better complement method for the continuous 1 day training data set. The voltage uses KNN Regression K=140, the current uses SVR cost =30, epsilon 0.3, the power factor uses SVR cost =60, epsilon 0.1, the power consumption uses X-12-ARIMA, and the immediate power consumption uses SVR cost =90. Epsilon 0.2 gets a better complement than the original LI.
