基于矩阵范数优化理论的用电数据质量提升算法  

Electricity Consumption Data Quality Improvement Algorithm Based on Matrix Norm Optimization Theory

作　　者：杨挺 孙兆帅 季浩[2] 叶芷杉 耿毅男 YANG Ting;SUN Zhaoshuai;JI Hao;YE Zhishan;GENG Yinan(School of Electrical and Information Engineering,Tianjin University,Nankai District,Tianjin 300072,China;State Grid Tianjin Electric Power Company,Hebei District,Tianjin 300010,China)

机构地区：[1]天津大学电气自动化与信息工程学院,天津市南开区300072 [2]国网天津市电力公司,天津市河北区300010

出　　处：《中国电机工程学报》2022年第10期3501-3511,共11页Proceedings of the CSEE

基　　金：国家重点研发计划项目(2017YFE0132100);国家自然科学基金项目(61971305);天津市自然科学基金重点项目(21JCZDJC00640)。

摘　　要：用电数据是智能电网大数据重要组成部分,也是基于人工智能方法进行负荷预测、需求响应以及台区线损治理和反窃电的基础样本数据来源。但用电信息采集设备工作环境复杂,用电数据缺失异常问题不可避免,严重影响数据驱动的效果。该文针对用电大数据存在的数据缺失、异常噪声等低质量问题,提出一种基于多范数优化的用电数据质量提升新算法,其中针对数据缺失和稀疏脉冲等多种现场采集噪声,采用核范数/1-范数/F-范数优化的低秩矩阵恢复模型和交替方向乘子算法求解,实现缺失数据恢复和异常噪声滤除,提高用电数据质量。所提方法具有不需要先验知识的训练,计算复杂度低的优势。算例结果表明,该文方法可以提高缺失数据恢复精度、改善数据质量,并且通过基于人工智能长短期记忆神经网络(long short term memory,LSTM)方法的短期负荷预测实验证明其可有效提高预测精度,对电力系统基于数据驱动的新兴高级应用具有良好的实际意义。Electricity consumption data is an important part of smart grid big data,and it is also a basic sample data source for load forecasting,demand response,station line loss management and anti-electricity theft based on artificial intelligence methods.However,missing data and abnormal noise are inevitable,which seriously affect the data-driven effect.Focusing on the low-quality problems such as missing data and abnormal noise in electricity consumption big data,a new algorithm for improving the quality of power consumption data based on multi-norm optimization was proposed.A low-rank matrix recovery model optimized by nuclear norm/1-norm/F-norm was used for missing data and various field acquisition noises such as sparse pulses.An alternating direction multiplier method was used to solve the problem,achieve missing data recovery and abnormal noise filtering,and improve the quality of electricity data.The proposed method has the advantages of not requiring prior knowledge training and low computational complexity.The results of calculation examples show that the method in this paper can improve the accuracy of missing data recovery and improve data quality.The short-term load forecasting experiment based on artificial intelligence long short term memory(LSTM)method proves that it can effectively improve the forecasting accuracy,and has practical significance for new and advanced applications based on data-driven power systems.

关 键 词：用电数据 聚类 多范数优化 交替方向乘子法 短期负荷预测 

分 类 号：TM71[电气工程—电力系统及自动化]