不同时间尺度下日用电量即时学习差异化模型  

Real-time learning differentiation model of daily electricity consumption under different time scales

作　　者：屈志坚 奚增辉 姚嵘 瞿海妮 陆嘉铭 QU Zhijian;XI Zenghui;YAO Rong;QU Haini;LU Jiaming(State Grid Shanghai Electric Power Company,Shanghai 200122,China)

机构地区：[1]国网上海市电力公司,上海200122

出　　处：《电子设计工程》2024年第15期21-24,29,共5页Electronic Design Engineering

基　　金：上海电力人工智能工程技术研究中心研究项目(19DZ2252800)。

摘　　要：日用电量分析受到异常数据的影响,导致分析结果与实际结果不一致。为了解决该问题,在不同时间尺度下构建日用电量即时学习差异化模型。对不同时间尺度下日用电量进行离群点的检测,剔除异常值,使用SPSS统计软件检测日用电量分布情况,获取日用电量正态分布特征。划分不同区域权重子区域,通过类间最近邻局部子域映射到高维特征空间中,计算正态分布特征关联系数,取正态分布特征。聚类处理服从正态分布的特征数据,用灰色E形关联度来评价多因子之间关系强度,采用特征量聚类方法构建即时学习差异化模型。由实验结果可知,该模型住宅区与工作区日用电量值与实际测得数据一致,最高值分别为125 kW·h和350 kW·h,具有精准的分析效果。The analysis of daily power consumption is affected by abnormal data,resulting in inconsistency between the analysis results and the actual results.In order to solve this problem,the real-time learning differentiation model of daily electricity consumption is constructed under different time scales.Detect outliers of daily power consumption at different time scales,eliminate outliers,use SPSS statistical software to detect the distribution of daily power consumption,and obtain the normal distribution characteristics of daily power consumption.Divide weight sub regions of different regions,map the nearest local sub regions between classes into high-dimensional feature space,calculate the correlation coefficient of normal distribution features,and take the normal distribution features.Clustering deals with the characteristic data of normal distribution,uses the grey E-shape correlation degree to evaluate the relationship strength between multiple factors,and uses the characteristic quantity clustering method to build the real-time learning differentiation model.It can be seen from the experimental results that the daily power consumption values of the residential area and the working area of the model are consistent with the actual measured data,with the highest values of 125 kW·h and 350 kW·h respectively,which has accurate analysis effect.

关 键 词：时间尺度 日用电量 即时学习 差异化模型 正态分布 

分 类 号：TN07[电子电信—物理电子学]