Stacking相异模型融合的实验室异常用电行为检测  

Laboratory Anomaly Detection of Electricity Consumption Behavior Based on Stacking Integrated Structures with Different Models

作　　者：陈静 王铭海 江灏[1] 缪希仁[1] 陈熙 郑垂锭 CHEN Jing;WANG Minghai;JIANG Hao;MIAO Xiren;CHEN Xi;ZHENG Chuiding(College of Electrical Engineering and Automation,Fuzhou University,Fuzhou 350108,China)

机构地区：[1]福州大学电气工程与自动化学院,福州350108

出　　处：《实验室研究与探索》2024年第1期231-237,共7页Research and Exploration In Laboratory

基　　金：福建省自然科学基金项目(2022J01566)。

摘　　要：针对当前高校实验室异常用电行为,提出一种基于Stacking相异模型融合的异常行为检测方法。考虑相异基学习器挖掘实验室用电行为规律的差异性,对相异基学习器进行优选。利用随机森林作为元学习器,充分融合相异基学习器的优势,弥补各基学习器的缺陷,构建基于Stacking相异模型融合的集成学习模型。通过算例对比分析,验证了基于Stacking相异模型融合的集成学习模型能有效提升单一分类器的异常检测效果,在准确率、F1分数、ROC曲线下面积和误检率上均优于Bagging、Voting、Adaboost等集成学习方法并能适应样本不平衡的情况。Aiming at the current abnormal electricity consumption behavior in university laboratories,this paper proposes a power anomaly detection method based on Stacking integrated with heterogeneous models.By consid-ering the diversity of the behavioral patterns of electricity usage in the laboratory,the heterogeneous base learners are selected based on their differences.Then,random forest is used as the meta-learner to fully integrate the ad-vantages of the heterogeneous base learners and compensate for their deficiencies,an integrated learning model is constructed based on Stacking heterogeneous model fusion.Finally,through comparative analysis of examples,the results show that the integrated learning model based on Stacking heterogeneous model fusion can effectively improve the classification performance of a single classifier.It outperforms other integrated learning methods such as Bagging,Voting,and Adaboost in terms of accuracy,F1 score,area under ROC curve,and false positive rate,and can adapt to imbalanced sample situations.

关 键 词：异常用电行为 Stacking结合策略 集成学习 实验室安全 

分 类 号：TM714[电气工程—电力系统及自动化]