VMD-Stacking集成学习的多特征变量短期负荷预测模型  

Multi-feature short-term load forecasting based onVMD-Stacking ensemble learning

作　　者：王士彬 何鑫 余成波 张未 陈佳 WANG Shibin;HE Xin;YU Chengbo;ZHANG Wei;CHEN Jia(State Grid Chongqing Shinan Electric Power Supply Branch,Chongqing 401336,China;School of Electrical and Electronic Engineering,Chongqing University of Technology,Chongqing 400054,China;Chongqing Energy Internet Engineering Technology Research Center,Chongqing 400054,China)

机构地区：[1]国网重庆市电力公司市南供电分公司,重庆401336 [2]重庆理工大学电气与电子工程学院,重庆400054 [3]重庆市能源互联网工程技术研究中心,重庆400054

出　　处：《兵器装备工程学报》2024年第2期218-224,共7页Journal of Ordnance Equipment Engineering

基　　金：重庆市自然科学基金面上项目(cstc2021jcyj-msxmX0251);重庆市教育委员会科学技术研究计划(KJQN202101115,KJQN202201157);重庆理工大学研究生教育高质量发展行动计划资助成果(gzlcx20232039,gzlcx20233120)。

摘　　要：针对当前短期电力负荷预测结果准确度不够高的问题,提出一种由变分模态分解(variational modal decomposition, VMD)和Stacking集成学习框架组合的多特征变量短期负荷预测模型。在预测前使用VMD算法将负荷数据分解,然后加入对模型重要性较高的特征变量,再建立由轻量级梯度提升机(light gradient boosting machine, LightGBM)与极限梯度提升机(extreme gradient boosting, XGBoost)融合的Stacking集成学习预测模型,并比较不同天气情况下对预测模型准确度的影响。经实际算例对比验证表明:多特征的VMD-Stacking集成学习预测模型的误差较小。采用VMD算法分解历史负荷序列,分解后子模态分量的周期性体现了出来,让模型预测波动性较大的负荷时更容易;温度、天气、农历和节假日情况等影响负荷变化的关键因素有被考虑到,模型的准确度得以提高;Stacking集成学习模型对各算法取长补短,泛化能力增强,预测的准确度高于单一模型。Aiming at the problem that the accuracy of current short-term power load forecasting results is not high enough,a multi-feature variable short-term load forecasting model is proposed,which is composed of variational modal decomposition(VMD)and Stacking ensemble learning framework.Before forecasting,the VMD algorithm is used to decompose the load data,and then feature variables that are of high importance to the model are added.Then,the Stacking ensemble learning forecasting model,which is composed with the light gradient boosting machine(LightGBM)and extreme gradient boosting(XGBoost),and the impact of different weather conditions on the accuracy of the forecasting model is compared.The comparison of actual examples shows that the error of multi feature VMD Stacking ensemble learning prediction model is small.Using VMD algorithm to decompose historical load series,the periodicity of sub modal components after decomposition is reflected,making it easier for the model to predict loads with high volatility;The key factors affecting load changes(such as temperature,weather,lunar calendar,and holiday conditions)have been taken into account,and the accuracy of the model has been improved;Stacking ensemble learning model learns from each other’s strong points to complement each other’s weak points,and its generalization ability is enhanced.Its prediction accuracy is higher than that of a single model.

关 键 词：短期电力负荷预测 变分模态分解 Stacking集成学习 多特征变量 轻量级梯度提升机 极限梯度提升机 

分 类 号：TM715[电气工程—电力系统及自动化]