智慧型非侵入式負載監測系統之設計

Development of an Intelligent Non-Intrusive Load Monitoring System

楊宏澤(H. T. Yang) ； 章學賢(Johnson H. H. Chang) ； 黃弘育(H. Y. Huang)

《中原學報》 30卷3期 (2002/09) Pp. 421-428

https://doi.org/10.6358/JCYU.200209.0421

摘要

在配電系統中，非侵入式負載監測系統是對能源消耗與電力品質獲得正確的資訊既方便且經濟的方法。本篇論文中發展一透過非侵入式的智慧型量測技術自動的對負載做監測，由於非侵入式的量測方式使得量測所需的儀器可以大為的減少。
爲了正確地了解連線負載的使用時間及辨認負載，所以從監測點量得的負載穩態變化及暫態的取樣信號應即時收集與分析。這些信號包含負載的實功、虛功、電壓與電流及其諧波成分等，經由離散小波轉換的方式對於負載的暫態初始狀態做偵測，除此之外利用離散小波轉換係數的量也可以用來確認負載的特徵。在這些監測到可能的暫態及穩態的特徵中，基因演算法的設計可以對於辨認的負載做最有效的特徵選取。基於所選取的特徵中，模糊學習向量量化法可以藉由個別負載的取樣信號建立一訓練的網路，在非侵入式負載監測系統中扮演著重要的角色，最後以現場實測與使用電磁暫態程式模擬做爲評估所發展的非侵入負載監測系統的效能。

關鍵字

非侵入式負載監測系統 ； 基因演算法 ； 負載辨識 ； 學習向量量化網路 ； 小波轉換

並列摘要

In a distribution system or an industrial plant, the non-intrusive load monitoring(NILM)system is a convenient and economical means of acquiring accurate energy consumption data and power quality information. This paper develops an NILM system using intelligent techniques to automatically monitor the loads that are being connected to and disconnected from the line through non-intrusive measurement. Due to non-intrusive measurement, the needed number of measurement instruments can be greatly reduced.
To accurately capture the timing of load connection and recognize the loads, variations of the steady state as well as the transient signals sampled from the monitoring point should be collected and analyzed on-line and real time. The signals analyzed include those of the active, reactive power, harmonics, voltage and current, etc. The signals of the voltage and current are further transformed through the discrete wavelet transformation(DWT)method and used to detect the transient signal initiation, which represents state variation of the loading. Besides, information on scales of the DWT coefficients is also used as features of recognition of the loads on line in addition to the steady state signals. Among the possible features of both steady state and transient signals monitored, the genetic algorithm is designed to select the most effective ones for recognized the loads. Based on the selected features, the fuzzy learning vector quantization(FLVQ) network is established and trained using the respective load sample signals. The FLVQ network then serves as the core of the NILM system for recognizing the loads as soon as the transient variation detected by the DWT. To verify the performance of the developed NILM system, the data for both the field tests and the simulations using the electromagnetic transient program(EMTP)are employed. The results, regarding the features selected by the GA, the load variation detected by the DWT, and the load recognized by the LVQ network, are presented in this paper.

並列關鍵字

Non-intrusive load monitoring system NILMS ； Genetic algorithm ； Load recognition ； LVQ network ； Wavelet transform