基于电力数据挖掘的涉污企业用电量预测方法研究  

Research on polluting enterprise electricity consumption forecasting method based on electric power data mining

作　　者：成贵学 乔臻 滕予非 唐伟 CHENG Guixue;QIAO Zhen;TENG Yufei;TANG Wei(College of Computer Science and Technology,Shanghai University of Electric Power,Shanghai 200090,China;State Grid Sichuan Electric Power Company Electric Power Research Institute,Chengdu 610072,China)

机构地区：[1]上海电力大学计算机科学与技术学院,上海200090 [2]国网四川省电力公司电力科学研究院,四川成都610072

出　　处：《现代电子技术》2022年第15期151-156,共6页Modern Electronics Technique

基　　金：国网四川省电力公司科技项目(52199718001A)。

摘　　要：为对工业涉污企业进行准确管控,提出一种基于用电特性聚类与ConvLSTM神经网络算法结合的涉污企业用电量预测方法。对于企业用电数据中的数据异常与缺失的问题,采用局部离群因子算法(LOF)筛选异常值后输入至灰色模型中进行校正;通过K⁃means算法对修正后的企业历史用电数据进行特征提取并分析其用电特征,考虑影响用电量的因素不仅包括日期特性、节日特性,还提取了重污染天气下政府对涉污企业的管控特性;构建ConvLSTM模型,充分挖掘企业数据时序性特征,有效提高涉污企业短期用电量预测精度。选择四川省成都市涉污企业的用电数据验证模型算法的有效性。验证结果表明,所提方法对于不同企业、不同类型日期均更有效,能更精确地预测企业未来用电的趋势。To control industrial polluting enterprises accurately,a polluting enterprise electricity consumption forecasting method based on the combination of power characteristic clustering and ConvLSTM neural network algorithm is proposed.For the abnormal and missing data in the enterprise′electricity consumption data,the local outlier factor(LOF)algorithm is used to screen the abnormal values and then input them into the gray model for correction.The K⁃means algorithm is used to extract features from the revised historical electricity consumption data of the enterprise,and its electricity consumption features are analyzed.In view of the factors that affect electricity consumption,including date characteristics and holiday characteristics,the government′s control characteristics of the polluting enterprise in heavy pollution weather is extracted.The ConvLSTM model is built to mine the time⁃series feature of enterprise data fully and improve the prediction accuracy of short⁃term electricity consumption effectively.The electricity consumption data of polluting enterprises in Chengdu,Sichuan Province are selected to verify the effectiveness of the model algorithm.The verification results show that the proposed method is more effective for different types of dates in different enterprises,and can predict the future power consumption trend of enterprises more accurately.

关 键 词：企业用电量预测 ConvLSTM LOF 灰色模型 K⁃means聚类算法 时序性特征 用电特性聚类 

分 类 号：TN911.1-34[电子电信—通信与信息系统] TP391.9[电子电信—信息与通信工程]