基于生成对抗和双重语义感知的配电网量测数据缺失重构

传统的数据缺失重构技术大多依赖数理统计方法和先验知识结合机理分析构建数学模型，但是配电网量测数据具有高维、时变、非线性特征，复杂度高、表征难度大，难以保证高精度重构。文中提出一种利用无监督生成对抗训练方式自主提取数据特征并结合双重语义感知重构约束实现数据缺失重构的方法。其中，基于二维卷积的重构模型和量测数据二维灰度图像化训练增强了模型泛化能力和稳定性。该方法无需先验知识的分布假设与显式物理建模，在保证数据特征提取最大化的同时，有效提高了重构数据的精确性。最后，利用实测数据验证了该方法在重构缺失数据上的有效性。

Reconstruction of Missing Measurement Data in Distribution Network Based on Generative Adversarial Network and Double Semantic Perception

Traditional data missing reconstruction technology mostly relies on mathematical statistics method and prior knowledge combined with mechanism analysis to construct mathematical models. However, measurement data in distribution network has high dimensional, time-variant, non-linear characteristics with high complexity and difficult characterization, and it is difficult to ensure high-precision reconstruction. In this paper, an unsupervised generation antagonism training method is proposed to extract data features autonomously and reconstruct missing data with reconstruction constraints of dual semantic perception. In the method, the reconstruction model based on two-dimensional convolution and the two-dimensional gray image training of measurement data enhance the generalization ability and stability of the model. This method does not need the distribution hypothesis of prior knowledge and explicit physical modeling, and it effectively improves the accuracy of reconstructed data while guaranteeing maximum feature extraction. Finally, the validity of this method in reconstructing missing data is verified by the field measured data.