基于改进最近邻算法的电力数据缺失处理方法  

Missing processing method for power data based on improved K-Nearest Neighbor algorithm

作　　者：江疆 杨秋勇 苏华权 JIANG Jiang;YANG Qiuyong;SU Huaquan(Guangdong Power Grid Co.,Ltd.,Gangzhou 510030,China)

机构地区：[1]广东电网有限责任公司,广东广州510030

出　　处：《电子设计工程》2025年第4期165-169,共5页Electronic Design Engineering

基　　金：南方电网公司科技项目(037800KK52200001,GDKJXM20200385)。

摘　　要：针对电力数据采集缺失算法存在的准确率低、计算开销大等缺陷,文中将最近邻算法与对抗神经网络结合,提出了一种电网用户信息缺失处理算法。对于最近邻算法存在的高维度数据处理能力差的问题,使用自编码器对高维数据进行降维,同时通过变分方法引入了隐变量学习数据中可连续、可解释的特征。利用最近邻算法生成数据样本标签,由对抗神经网络根据样本标签和自编码器输出特征最终生成缺失数据。在公开数据集进行的实验测试中,所提算法的准确率与迭代次数在所有应用场景及对比算法中均为最优,充分表明了算法的高效性和工程实用性。In response to the shortcomings of low accuracy and high computational overhead in missing algorithms for power data collection,this paper proposes a power grid user information missing processing algorithm by combining the K-Nearest Neighbor algorithm with adversarial neural networks.For the problem of poor processing ability of high-dimensional data in the K-Nearest Neighbor algorithm,the Autoencoder is used to reduce the dimension of high-dimensional data,and the variational method is used to introduce the continuous and interpretable features in the implicit variable learning data.The K-Nearest Neighbor algorithm is used to generate data sample labels,and the countermeasures neural network generates missing data according to the sample labels and Autoencoder output characteristics.In experimental tests conducted on publicly available datasets,the accuracy and iteration times of the proposed algorithm were the best in all application scenarios and comparison algorithms,fully demonstrating the efficiency and engineering practicality of the algorithm.

关 键 词：最近邻算法 变分自编码器 对抗神经网络 数据缺失处理 数据分析 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TN929.5[自动化与计算机技术—计算机科学与技术]