基于改进生成对抗网络的台区采集数据修复

Missing Data Imputation in Transformer District Based on Improved Generative Adversarial Network

低压配电网台区位于输配电系统的末端,是开展配电系统管控的基础环节。受不可抗力的影响,台区终端采集数据普遍存在缺失值,整体数据质量较差,进而影响信息的正确性和决策分析的准确度。传统的数据修复方法忽略了台区数据的周期性和时序性,修复精度较低。该文提出一种基于生成对抗网络(generative adversarial network,GAN)的配电网台区缺失采集数据修复模型,改进了GAN网络的结构,为判别器额外设计了提示机制,使其能够尽可能地利用未缺失信息,潜在地拟合原始数据的分布特征。所提出的方法不需要利用完整的数据集进行训练,整体运行在无监督的环境下,更适用于复杂的生产实际,实验结果表明,所提方法能够高精度地对台区缺失数据进行修复。

The low-voltage transformer districts, located at the end of power system, is the key point of the distribution network management and control. Affected by different factors, missing data is common in the district teminals, and the overall data quality is poor, which in turn affects the information accuracy and decision analysis. The traditional method of data imputation ignores the periodicity and temporality of district data and the imputation accuracy is relatively low. This paper proposes a model for imputating missing data in transformer district terminals based on the generative adversarial network, and a hint mechanism is additionally designed for the discriminator. The structure of GAN is improved to make it possible to use as much unmissing information as possible to potentially fit the distribution characteristics of the original data. The proposed method does not need to use a complete data set for training. It runs in an unsupervised environment, more suitable for practices. The experimental results show that the proposed method can impute the missing data with high precision.