Non-intrusive load monitoring using artificial intelligence classifiers: Performance analysis of machine learning techniques
Non-intrusive load monitoring using artificial intelligence classifiers: Performance analysis of machine learning techniques
复制标题
DOI：
10.1016/j.epsr.2021.107347
复制
发表时间：
2021-05-15
影响因子：
3.9
通讯作者：
Poma, C. E. P.
中科院分区：
文献类型：
--
作者：
Monteiro, R. V. A.;de Santana, J. C. R.;Poma, C. E. P.
近年来，已经提出了用于负载监测的策略以减轻功耗。在几项报告的研究中已经发现，随着向消费者提供更多关于他们的电力消耗的信息，将发生更多的电力节能。这样，非侵入式负载监测（NILM）已被研究和应用在现实生活中的应用。它包括通过仅在住宅消费者的一个位置测量电信号来检测和分类电器开/关状态。已经使用不同的技术进行了几项研究，以提高该策略的准确性。在本文中，电磁暂态考虑到，先进的人工分类器之间的性能分析。已经发现，1D卷积神经网络在这种情况下表现更好，并且电流信号更适合NILM，一旦它比电压和功率信号携带更多的特征。
In recent years, strategies for load monitoring have been proposed to mitigate power consumption. It has been found, in several reported studies, that as more information is provided for consumers about their electricity consumption, more power energy conservation will occur. In this way, Non-Intrusive Load Monitoring (NILM) has been studied and applied in real-life applications. It consists of detecting and classifying appliances on/off states by measuring electrical signals only at one location of the residential consumer. Several studies have been made using different techniques to improve the accuracy of this strategy. In this paper electromagnetic transients are taking into account and, a performance analysis between cutting-edge artificial classifiers is made. It has been found that 1D convolutional neural networks perform better for this case and electrical current signals are more suitable for NILM, once it carries more features than voltage and power signals.