基于XGBoost与Stacking模型融合的短期母线负荷预测  

Short-term bus load forecasting based on XGBoost and Stacking model fusion

作　　者：刘波 秦川[1] 鞠平[1] 赵静波 陈彦翔 赵健 LIU Bo;QIN Chuan;JU Ping;ZHAO Jingbo;CHEN Yanxiang;ZHAO Jian(College of Energy and Electrical Engineering,Hohai University,Nanjing 211100,China;State Grid Jiangsu Electric Power Company Research Institute,Nanjing 210008,China;State Grid Nanjing Power Supply Company,Nanjing 210019,China)

机构地区：[1]河海大学能源与电气学院,江苏南京211100 [2]国网江苏省电力有限公司电力科学研究院,江苏南京210008 [3]国网南京供电公司,江苏南京210019

出　　处：《电力自动化设备》2020年第3期147-153,共7页Electric Power Automation Equipment

基　　金：国家自然科学基金重点资助项目(51837004);“111”计划(新能源发电与智能电网学科创新引智基地)(B14022)。

摘　　要：母线负荷预测对于电网安全稳定调度具有重要意义,但母线负荷随机波动性较强,其负荷类型因供电区域的差异而不同。为此,提出一种基于极限梯度提升(XGBoost)与Stacking模型融合的短期母线负荷预测方法。基于XGBoost建立多个母线负荷预测元模型,组合构成Stacking模型融合的元模型层,连接一个XGBoost模型对元模型进行融合,整体构成综合预测系统,并采用粒子群优化算法优化系统参数。通过对具有不同负荷属性的220 kV母线进行实例分析,验证了所提方法的有效性与适用性。Bus load forecasting plays an important role in safe and stable dispatching of power grid,but bus load has strong stochastic fluctuation,and its attributes are various because of the difference of power supply areas,for which,a short-term bus load forecasting method based on XGBoost(eXtreme Gradient Boosting) and Stacking model fusion is proposed. Multiple bus load foresting meta-models are built based on XGBoost to form the meta-model layer of Stacking model fusion,an XGBoost model is connected to the meta-model layer for fusion,and the comprehensive forecasting system is formed. The particle swarm optimization algorithm is adopted to optimize the system parameters. The cases of 220 kV bus with different load attributes are analyzed,and the validity and applicability of the proposed method are verified.

关 键 词：母线负荷 XGBoost 元模型 Stacking模型融合 粒子群优化算法 

分 类 号：TM715[电气工程—电力系统及自动化]