基于加权宽度学习的异常用电辨识研究
Research on Abnormal Electricity Identification Based on Weighted Broad Learning System
姚影 1陆俊 1肖琦 1龚钢军 1徐志强 2辛培哲3
作者信息
- 1. 北京市能源电力信息安全工程技术研究中心(华北电力大学),北京市 昌平区 102206
- 2. 国网湖南省电力公司经济技术研究院,湖南省长沙市 410004
- 3. 国家电网北京经济技术研究院,北京市 昌平区 102211
- 折叠
摘要
针对异常用电与正常用电样本类别不平衡关系及现有模型训练耗时、缺乏可扩展性的问题,提出一种基于加权宽度学习(weighted broad learning system,WBLS)的异常用电辨识模型.首先,考虑到样本间类别不平衡关系,在目标函数中使用样本权重约束每个类对模型的贡献,样本权重根据样本分布情况个性化赋予,并通过岭回归广义逆高效地建立WBLS辨识模型.其次,基于新增加的用电样本数据,通过增量学习算法对模型进行更新和重构.实验结果表明该模型提高了对异常用电样本的辨识精度,并在增加用电样本的情况下,可以快速地对旧模型进行更新和扩展.
Abstract
Aiming at the problem of unbalanced relationship between abnormal electricity consumption and normal electricity consumption sample categories,time-consuming training and lack of scalability of existing models,an abnormal electricity consumption identification model based on Weighted Broad Learning System(WBLS)was proposed.Firstly,considering the class imbalance relationship between samples,the sample weight is used in the objective function to constrain the contribution of each class to the model,and the sample weight is personalized according to the distribution of samples,and the generalized inverse WBLS identification model is established efficiently by ridge regression.Secondly,based on the newly added electricity consumption sample data,the model is updated and reconstructed by the incremental learning algorithm.The experimental results show that the model improves the identification accuracy of abnormal electricity samples,and can quickly update and expand the old model with the increase of electricity samples.
关键词
异常用电/加权宽度学习/类不平衡/增量学习Key words
abnormal power consumption/weighted broad learning system/class imbalance/incremental learning引用本文复制引用
基金项目
国家电网总部管理科技项目(5700-2-2252219 A-1-1-ZN)
出版年
2024
国家科技期刊平台
NSTL
维普
万方数据