基于时序信号图像编码和生成对抗网络的配电网台区数据修复

受不可抗力影响，配电网低压台区数据中普遍存在缺失值，整体数据质量较差，限制了台区的精益化管理水平。传统的数据修复方法忽略了数据的周期性和时序性，修复精度较低。提出了一种基于图像编码和生成对抗网络的台区缺失数据修复方法。首先引入了一种一维时序信号编码图像预处理方法，将原始的时序信号转换为格拉姆角场图像，然后利用卷积神经网络在图像特征提取上的强大优势构建了生成对抗网络模型。结合像素损失和相似性损失的双重约束条件增强了生成图像的质量。整体流程由数据驱动，无需先验知识的分布假设与显式物理建模。最后的算例结果表明，该方法能够较为精确地实现台区缺失数据的修复。

Missing data imputation in a transformer district based on time series imaging encoding and a generative adversarial network

For various reasons, there are generally missing values in low voltage transformer districts, and overall data quality is poor. This limits lean management for a distribution network. Traditional data imputation methods ignore the periodicity and temporality of data and the imputation accuracy is relatively low. This paper proposes an imputation method based on imaging encoding and a generative adversarial network. First, a method for one-dimensional time series encoding to a two-dimension image is introduced, the original signals are transformed to Gramian angular field images. Second, using the powerful advantages of a convolutional neural network in image feature extraction, a generative adversarial network model is constructed. Combining the dual constraints of pixel and similarity loss, the quality of generated images is improved. The overall process is driven by data, does not need the distribution hypothesis of prior knowledge and explicit physical modeling. Lastly, experimental results show that this method could accurately impute missing values in a transformer district. This work is supported by the Science and Technology Project of State Grid Corporation of China (No. 52020116000G).