| 引用本文: | 涂京,周明,宋旭帆,周光东,李庚银.基于监督学习的非侵入式负荷监测算法比较[J].电力自动化设备,2018,(12): |
| TU Jing,ZHOU Ming,SONG Xufan,ZHOU Guangdong,LI Gengyin.Comparison of supervised learning-based non-intrusive load monitoring algorithms[J].Electric Power Automation Equipment,2018,(12): |
|
| 摘要: |
| 非侵入式负荷监测(NILM)能够在不干扰用户正常用电的情况下，低成本地实现用户用电设备类型的识别和用电负荷的分解，因此非常适用于家庭用户用电监测。大量智能电表在家庭用户中的安装为居民NILM提供了数据支撑，也使得居民NILM研究成为热点。基于家庭负荷稳态电流样本，采用负荷电流谐波系数作为负荷分类特征，建立了基于多层感知器(MLP)神经网络、k-近邻算法、逻辑回归、支持向量机的4种NILM分类模型，利用BLUED数据库对4种分类器进行训练和测试，对比分析其在识别精度、训练时间、识别速度和抗噪能力方面的表现，并对其在家庭负荷识别中的应用效果进行对比研究。结果表明，4种分类器中MLP神经网络具有总体最优的分类效果和计算性能，更适用于家庭用户负荷监测。 |
| 关键词: 非侵入式负荷监测 负荷识别 多层感知器神经网络 支持向量机 监督学习 |
| DOI：10.16081/j.issn.1006-6047.2018.12.019 |
| 分类号:TM721;TM764 |
| 基金项目:国家重点研发计划项目(2016YFB0901100) |
|
| Comparison of supervised learning-based non-intrusive load monitoring algorithms |
|
TU Jing, ZHOU Ming, SONG Xufan, ZHOU Guangdong, LI Gengyin
|
|
State Key Laboratory for Alternate Electrical Power System with Renewable Energy Sources, North China Electric Power University, Beijing 102206, China
|
| Abstract: |
| NILM(Non-Intrusive Load Monitoring) can realize the type identification of the user's electrical equipment and the load decomposition with low costs and without disturbing the users' normal electricity consumption, which is especially suitable for residential load monitoring. The installations of massive smart meters for domestic consumers provide data support for residential NILM, which makes the residential NILM become a hot research topic. Based on the samples of residential load steady-state current, the harmonic coefficients of the load steady-state current are used as the load classification characteristics, and four NILM classification models based on MLP(Multi-Layer Perceptions) neural network, k-nearest neighbor algorithm, logistic regression and SVM(Support Vector Machine) are established. The four types of classifiers are trained and tested with the BLUED database, their performances of identification accuracy, training time, identification speed and anti-noise property are compared, and their application effects in residential load identification are contrasted. Results show that, among the four classifiers, the MLP neural network has the optimal classification effect and computational performance, which is more suitable for monitoring the residential load. |
| Key words: non-intrusive load monitoring load identification multi-layer perceptions neural networks support vector machines supervised learning |
