- 學位論文
基於一個新穎的特徵萃取方法及人工智慧技術之自適應非侵入式負載監測系統研發
A Development of Adaptive Nonintrusive Load Monitoring System based on a Novel Feature Extraction Method and Artificial Intelligent Technique
摘要
有別於傳統的侵入式負載監測系統，非侵入式負載監測(Non-Intrusive Load Monitoring, NILM)系統不需要在待監測的每一個負載上安裝感測器。而僅需要在電力供應入口處安裝感測器，藉由分析在電力入口處所量測的電壓與電流波形，監測系統即可得知各個負載的電力使用情形。 本文之非侵入式負載監測系統結合一個新式的特徵萃取方法及人工智慧辨識技術所設計而成，且此監測系統透過最佳化策略可以達到自適應的能力。本監測系統藉由使用k-最近鄰居法則(k-Nearest-Neighbor Rule, k-NNR)及倒傳遞類神經網路(Back-Propagation Artificial Neural Network, BP-ANN)以辨識實際負載的啟動(start-up)及停止(shut-down)。經由在不同的實測環境下執行負載辨識，本論文所研發的監測系統的總辨識率都可以達到88.30%以上(最高為100%)，且藉由採用人工免疫系統(Artificial Immune System, AIS)，具備自適應的能力。透過實測辨識結果，足以證明所設計的監測系統具有可行性、正確性及強健性。
並列摘要
Traditional load monitoring system requires installations of sensors installed at different loads in order to monitor their operational states. On the other hand, the nonintrusive load monitoring (NILM) system requires installing a current or voltage sensor at the main electrical panel. By analyzing the voltage and current signals, the power usage of each load can be obtained. In this thesis, an adaptive nonintrusive load monitoring system with an optimization strategy that integrates a new feature extraction method and artificial intelligence recognition technique is proposed. The proposed system is used to monitoring the energizing and de-energizing state of each load by applying k-nearest-neighbor rules and back-propagation artificial neural networks. Through the experimental tests in different environments, the worst overall correct rate for all scenarios is 88.30%. Additionally, the proposed system can achieve the ability of adaptation by applying artificial immune system. These results show that the proposed system has feasibility, accuracy, and robustness.