基于特征选择和机器学习的台区线损计算方法  

A line loss calculation method based on feature selection and machine learning algorithm

作　　者：刘度度 任浪 肖坤 吴邦飞 颜钟宗 温和[2] LIU Dudu;REN Lang;XIAO Kun;WU Bangfei;YAN Zhongzong;WEN He(Zhangjiajie Power Supply Branch,State Grid Hunan Electric Power Company,Zhangjiajie 427000,Hunan,China;School of Electrical and Information Engineering,Hunan University,Changsha 410082,China)

机构地区：[1]国网湖南省电力公司张家界供电分公司,湖南张家界427000 [2]湖南大学电气与信息工程学院,长沙410082

出　　处：《电测与仪表》2024年第12期133-143,共11页Electrical Measurement & Instrumentation

基　　金：国网湖南省电力有限公司科技项目(5216G02100FF)。

摘　　要：降低电网损耗是节能减排的重要技术措施,线损率计算是电网企业制定降损目标和实现碳排放趋势预测的重要途径。现有线损计算研究主要关注模型的构建,忽视特征分析问题。基于此,文章提出了一种基于LightGBM(light gradient boosting machine)模型的台区线损率计算方法。分析了电气特征指标选取问题,通过探索台区电气特征指标分布及与线损率的关联关系确立模型输入。根据特征工程结果,建立基于LightGBM模型的台区线损率计算模型,揭示了不同模型参数和电气特征指标输入下对模型计算结果的影响。通过某市8000余个台区的历史数据验证方法的有效性。实验结果表明,所提方法的MSE(mean-square error)和MAPE(mean absolute percentage error)可分别达到0.020和2.459%,对比现有相关研究方法具有良好的计算精度。The power grid loss reduction is an important technical measure for energy conservation and emission reduction.And line loss rate calculation is an important way for electric utilities to formulate loss reduction targets and forecast the carbon emission.The existing research on line loss calculation mainly focuses on the construction of the model,ignoring the issues of feature analysis.To this end,this paper proposes a line loss calculation method in the station area based on the LightGBM.The selection of electrical features is analyzed,and then,the model input is established by exploring the distribution of electrical feature index and its correlation with the line loss rate.According to the results of feature engineering,a line loss rate calculation model is established based on the LightGBM,and the influence of different model parameters and electrical features on model calculation results is revealed.The validity of the proposed method is verified through the historical data of more than 8,000 low-voltage station areas in a city.The experimental results show that,compared with the existing related research methods,the proposed method has 0.020 MSE and 2.459%MAPE respectively.

关 键 词：线损计算 特征选择 机器学习 台区 LightGBM 

分 类 号：TM727[电气工程—电力系统及自动化]