摘要
窃电问题伴随着我国电力事业的发展,由常规窃电转向高科技窃电,呈现多样化、隐蔽化、高科技化等特征。针对该问题,对用户用电历史数据分析,提取特征指标进行降维并结合局部离群因子、极限学习机算法对窃电行为进行检测。对用户的负荷数据进行分类,提出度量负荷曲线的4种指标并得出13个特征变量。在分类的基础上对提取出的特征变量进行降维,利用局部离群因子筛选出用电异常用户。采用8个窃电判别指标并提取主成分,将用电异常用户提取主成分后的窃电判别指标数据作为训练样本输入模型。实验结果为训练样本与测试样本的预测正确率分别达到99. 55%和98. 67%。实验证明该模型对窃电用户有很好的识别效果。
With the development of China s electric power industry, the problem of stealing electricity has changed from conventional stealing electricity to high-tech stealing electricity, showing the characteristics of diversification, concealment and hi-tech. In order to solve this problem, we analyzed the user s electrical historical data, extracted the feature index to reduce the dimension, and combined the local outlier factor with the extreme learning machine to detect the electricity stealing behavior. The load data of users were classified. Four indexes of load curve were proposed and 13 characteristic variables were obtained. On the basis of the classification, the extracted feature variables were dimensionality reduced. The local outlier factors were used to screen out the users with electrical anomalies. We adopted eight criteria for stealing electricity and extracted principal components. The indicator data of electricity abnormal users after extracting principal components were taken as input models of training samples. The experiment results show that the prediction accuracy of training samples and test samples reaches 99.55% and 98.67% respectively, and the model has a good recognition effect for users with electricity theft.
作者
李梓欣
李川
李英娜
Li Zixin;Li Chuan;Li Yingna(School of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500,Yunnan,China)
出处
《计算机应用与软件》
北大核心
2018年第12期179-186,共8页
Computer Applications and Software
基金
国家自然科学基金项目(51567013)
关键词
窃电检测
聚类分析
特征提取
主成分分析
局部离群因子
极限学习机
Electricity larceny detection
Clustering analysis
Feature extraction
Principal component analysis Local outlier factor
Extreme learning machine
