本文利用智慧電錶為基礎,製作一嵌入式系統並能有效辨識正在使用之負載,繼而即時監控家庭詳細之用電狀況。傳統侵入式監測系統意旨將電錶裝置在每個須要監測之負載上,如此耗費不貲,且一旦監測目標數量增加,不可避地免成本及訊息資料量亦相應增加。本研究使用非侵入式監測系統(Non-Intrusive Load Monitoring system, NILMsystem),其方式只須於電力入口處安裝一只智慧電錶,如此可降低施作成本及減少資料量之蒐集與傳遞,可提升拆裝及維護之便利性。然而此類NILM的技術核心在於分類演算法的建構,職是之故,我們利用所量測到的電力資料,萃取電力特徵值,經過負載辨識之處理,達到辨識監測系統下各負載使用情況。 本系統提供兩類功能:若能先行觀察各家電之用電特徵值並事先儲存至資料庫內,應用決策樹(Decision Tree)分類法,根據電力特徵值找出最佳分類方式,其分類後之各類群利用最鄰近搜尋法(Nearest Neighbor Search, NNS)準確辨識出為何種負載,則能即時辨識資料庫內究竟何種負載正在啟動或停止;若使用負載未在資料庫內,則系統亦能依據新負載之特徵值,根據決策樹分類結果告知使用者負載類型。以上兩者辨識功能,經過實際實驗測試系統辨識狀態成功率達90%以上。
In this research, an embedded system, which can effectively identify the “on” or “off” status of the load during use, is created based on the use of smart meter. Subsequently, a detailed real-time monitoring of the electricity usage in a household can be available instantly. Traditionally, invasive monitoring systems are used to monitor the load by installing a metering device at the sockets of every appliance. However, this will lead to high cost. Moreover, once the number of monitored targets increase, it is inevitable that the costs and corresponding amount of resulting data increase as well. A Non-Intrusive Load Monitoring system, “NILM system” for short, is used for this research. The implementation of such system is as simple as fitting a smart electricity meter at the entrance of the power line. This not only reduces the installation cost and the amount of data collection and transmission, it also streamlines the installation, disassembly and maintenance process. However, the core of such NILM technology lies in electrical signal identification and the construction of the classification algorithms. Therefore, we extract the power characteristic values by collecting the measured power data. After the processing of the load identification, the monitoring of the load situation at every target is achieved by the identification monitoring system. This system offers 2 types of functions: If we can first observe the characteristic value of each appliance and store it in the data bank beforehand, the Decision Tree classification method will figure out the best qualified group according to the power characteristic values. After the classification process, the Nearest Neighbor Search,(NNS) method is used on the chosen group. The real-time data bank accurately identifies exactly what kind of load is being run or stopped. If the electricity load is getting underway and its characteristic value is not in the database, the system can still inform the user what type of load exists based on the feature values of the new load and the results of the Decision Tree classification. Both of the above identification functions, after being tested several times in actual experiments, demonstrate general success rates above 90% in load identification.