以特徵擷取及基因規劃為基礎之非侵入式負載監測系統設計

Feature Extraction and Genetic Programming Based Non-intrusive Load Monitoring System Design

李哲毅(Che-Yi Lee)

指導教授 ： 洪穎怡 ； 楊宏澤

中原大學/工學院/電機工程研究所/碩士(2009年)

https://doi.org/10.6840/cycu200900435

摘要

傳統的侵入式負載監測系統需要在所要監測的每一個負載上加裝硬體設備，而非侵入式負載監測系統僅需要於電力供給入口附近量測電力波形，經由系統作分析後，即可得知各個負載的上下線狀況。非侵入式負載監測系統不但安裝簡單，拆卸容易，花費低，但其所需的負載辨識技術卻較侵入式負載監測系統來的複雜。
本論文提出以特徵擷取及基因規劃為基礎之非侵入式負載監測方法，首先對所量測到的電流波形作特徵擷取以產生統計值指標，再將統計值指標作正歸化和取log值，處理過的統計值指標輸入至基因規劃作演化運算，以找出最佳解來產生新特徵值；另外，並將每一種負載組合的實功、虛功和電流諧波量所組成的特徵向量，送入至基因規劃作世代演化，以找出最佳解來產生新特徵值；將上述兩種方法所產生的新特徵值、統計值指標以及特徵向量分別輸入至倒傳遞類神經網路或支援向量機，以辨識包含固定負載及變動負載在內之不同負載組合。
從供給電壓不變動和±5%變動兩個案例的結果得知，於電壓±5%變動的情況下，藉由統計值指標和基因規劃所找到的解仍可獲得滿意的辨識效率。本文並意外發現，於所有方法中，以類神經網路辨識第一階特徵萃取出的統計值指標所得的辨識率為最佳，可驗證統計值指標有著優越的負載辨識能力。

關鍵字

非侵入式負載監測 ； 基因規劃 ； 特徵萃取 ； 類神經網路 ； 支援向量機

並列摘要

Traditional load monitoring system needs to install hardware circuit on each load to be monitored. However, non-intrusive load monitoring system (NILMS) only needs to install a monitoring device on the power source entrance to collect the power waveform data. By analyzing energy consumption of the power waveform, the system can identify the loads accordingly. Therefore, the monitoring facilities and cost needed by the NILMS is less than those of the conventional one. In addition, NILMS is also easier to install, remove and maintain. But the techniques needed for load identification are more sophisticated as compared with the convectional one.
A feature extraction and genetic programming (GP) based NILMS is proposed in this thesis. First the statistical indexes are generated by feature extraction from the raw current data recorded from the power source entrance. Then the statistical indexes will be normalized and taken logarithm of the elements of the normalized features. The process statistical indexes are inputted to GP in order to find the solution and generate new features. In addition, the feature vectors which are composed of real power, reactive power, and the values of current harmonics are inputted to GP in order to find the solution and generate new features. Two kinds of new features which are generated by the above-mentioned ways, statistical indexes and feature vector are inputted to artificial neural networks (ANN) and support vector machines (SVM) respectively for the identification of various load combinations which are composed of constant loads and variable loads.
From two cases—constant supply voltage value and variable supply voltage value within 5%, using GP and statistical indexes can still search the solution which can get not bad recognition rate. In this thesis, it seems use ANN to identify the statistical indexes which are generated by feature extraction can get the best recognition rate in all ways. It is found that the statistical indexes are verified to be the most efficient features for the load identification.

並列關鍵字

and support vector machines (SVM) ； feature extraction ； genetic programming (GP) ； non-intrusive load monitoring system (NILMS) ； artificial neural networks (ANN)

參考文獻

[2] S. Drenker and A. Kader, “Nonintrusive Monitoring of Electric Loads,” IEEE Computer Applications in Power, Vol. 12, no. 3, pp. 47-51, 1999.

[3] G. W. Hart, “Nonintrusive Appliance Load Monitoring,” Proceedings of the IEEE, Vol. 80, no. 12, pp. 1870-1891, Dec. 1992.

[4] G. W. Hart, “Residential Energy Monitoring And Computerized Surveillance Via Utility Power Flows,” IEEE Technology and Society Magazine, Vol. 8, no. 2, pp. 12 – 16, Jun. 1989.

[6] S. B. Leeb, S. R. Shaw, and J. L. Kirtley Jr., “Transient Event Detection in Spectral Envelope Estimates For Nonintrusive Load Monitoring,” IEEE Trans. on Power Delivery, Vol. 10, no. 3, pp. 1200–1210, Jul. 1995.

[7] K. D. Lee, S. B. Leeb, L. K. Norford, P. R. Armstrong, J. Holloway, and S. R. Shaw, “Estimation of Variable-Speed-Drive Power Consumption From Harmonic Content,” IEEE Trans. Energy Convers., Vol. 20, no. 3, pp. 566–574, Sep. 2005.