基于云端协同的家用电器负荷辨识能力提升技术研究

Research on technology of improving the identification ability of household electrical appliances based on cloud collaboration

中文关键词: 负荷辨识  云端协同  时空特征  K最近邻法

英文关键词: load identification  cloud collaboration  space⁃time characteristics  K⁃nearest neighbor

中文摘要:

      居民家用电器类型丰富，具有相似特征量的电器种类往往很多，给非介入辨识带来电器类型不确定、辨识准确度有待提升等问题，所以提出了一种基于多类型特征交互的云端协同的负荷辨识方法。首先端侧基于高频采样进行特征提取及负荷辨识，基于 CUSUM 事件检测方法提高检测过程中小偏移事件的检测灵敏度，应用轻量级邻近辨识方法进行基本电器辨识并将不确定电器的时空特征上送云端；其次云侧辨识能力提升，构建了包含固有特征、时空特征及统计特征的 16 维云侧历史特征库，提出了面向多维时空特征的最邻近原则优化辨识技术；最后构建云侧闭环升级机制，云侧将差异性特征回送端侧完善终端电器特征库，综合实现不确定电器细化识别能力提升。以南京某用户为例，云端协同相较于终端的居民一般电器的辨识率从 67%提升至 91%，并实现了无法辨识电器的辨识，有效验证了算法的有效性。

英文摘要:

      Residents have a wide variety of household electrical appliances with similar characteristics, which brings problems to non-intrusive identification such as uncertainty in the types of appliances and the need to improve the accuracy of identification.In response to this problem, a cloud-based collaborative load identification method based on multi - type feature interaction is proposed. Firstly, the end-side performs feature extraction and load identification based on high-frequency sampling, to improve the detection sensitivity of small offset events in the detection process based on the CUSUM event detection method, using the light-weight proximity recognition method to perform basic electrical appliance identification and upload the spatial-temporal characteristics of uncertain electrical appliances to the cloud. Secondly, the cloud side recognition ability is improved by constructing a 16-dimensional cloud-side historical feature database consisting of inherent features, spatio - temporal features and statistical features.An optimized identification technology based on the nearest neigh-bor principle for multi-dimensional spatio-temporal features is proposed. Finally, a closed - loop cloud upgrade mechanism is built.The cloud side sends back the different characteristics to the terminal to improve the terminal electrical feature library, comprehensively realizing the improvement of the ability to identify uncertain electrical appliances. Taking a user in Nanjing province as an example, the recognition rate of cloud collaboration has increased from 67% to 91% compared with that of the terminal. The recognition rate of unrecognized appliances has been realized, effectively verifying the effectiveness of the algorithm.