面向配用电侧负荷数据的深度端到端超分辨率感知方法

Deep End-to-end Super-resolution Perception Method for Load Data at Distribution Side

在智能电网中,精准的数据采集是整个系统安全与经济运行的基础。随着信息与物理系统融合的不断加深,各类大数据应用与实时控制等任务对采集高频数据的要求不断提高。然而,提高数据采样频率必然给系统造成更高的数据通信与存储负担。文中提出了一种基于深度学习的超分辨率感知技术,用于从低频采样的传感器数据中恢复精确的高频数据。具体地,提出了一种基于门控循环单元网络的深度端到端超分辨率感知方法,包括特征提取、关系推断、信息重建3个部分。特征提取部分采用一维卷积网络对低频数据进行特征提取;在关系推断部分应用门控循环单元网络对获得的特征进行学习,推断低频数据同高频数据间内在关系;在信息重建部分则使用全连接层对推断信息进行重建,获得对应的高频数据。采用所提方法对居民用户和工业用户功率数据及输电线电压数据进行超分辨率感知,同时使用恢复的高频数据进行负荷状态识别。算例结果表明所提方法能够精准有效地恢复低频数据的丢失信息并提升负荷识别等实际应用的准确性。

In smart grid,accurate data acquisition is the basis of the whole system security and economic operation.With the deepening integration of cyber-physical systems,the requirements of various big data applications and real-time control tasks for the acquisition of high-frequency data are keeping rising.However,increasing the data sampling frequency will inevitably cause higher data communication and storage burden to the system.In this paper,a deep learning based super-resolution perception technology is proposed to recover accurate high-frequency data from low-frequency sampled sensor data.Specifically,a deep endto-end super-resolution perception method based on the gated recurrent unit(GRU)network is proposed,which includes feature extraction,relationship inference and information reconstruction.In the feature extraction part,an one-dimensional convolution network is used to extract the features of low-frequency data.In the relation inference part,the GRU network is used to learn the obtained features to infer the internal relationship between the low-frequency data and the high-frequency data.In the information reconstruction part,the inferential information is reconstructed with the full connection layer to obtain the corresponding highfrequency data.The proposed method is used for super-resolution perception of the power data of residential and industrial users and the voltage data of transmission lines,and the recovered high-frequency data is used for load state identification.The example results show that the proposed method can accurately and effectively recover the lost information of low-frequency data and improve the accuracy of practical applications such as load identification.