基于集成学习的多模型融合光伏发电功率动态修正预测  

Multi-model fusion dynamically correct prediction of PV power generation based on ensemble learning

作　　者：忻俊杰 徐良 李永杰 李文磊[1] XIN Junjie;XU Liang;LI Yongjie;LI Wenlei(Faculty of Electrical Engineering and Computer Science,Ningbo University,Ningbo 315211,China)

机构地区：[1]宁波大学信息科学与工程学院,浙江宁波315211

出　　处：《传感器与微系统》2021年第4期117-121,共5页Transducer and Microsystem Technologies

基　　金：浙江省自然科学基金资助项目(LY18F03004,LQ20E070001);宁波市自然科学基金资助项目(2019A610126,2019A610119)。

摘　　要：将XGBoost,Light GBM与NGBoost三种集成学习模型进行动态权值融合以保留特征相关性,并引入修正系数来减小误差累积对预测准确性的影响。最终模型对澳大利亚中部康奈伦机场位置的光伏阵列的发电功率按月进行了预测,并引入多层感知器、支持向量回归与长短期记忆网络进行对比实验。预测结果表明:基于集成学习模型融合的光伏(PV)发电动态修正预测要显著优于其他预测模型,在平均绝对误差方面提升幅度为10%~19%,较之仅使用权值融合的集成学习模型提升幅度为14%。Dynamic weight fusion of three integrated learning models of XGBoost,Light GBM and NGBoost is carried out to retain feature correlation and correction coefficient is introduced to reduce the effect of accumulation of errors on prediction accuracy.On this basis,the final fusion model is used to predict the photovoltaic power generation of the photovoltaic array at the location of Connellan Airport in central Australia on a monthly frequency.At the same time,three kinds of machine learning regression models,multilayer perceptron,support vector regression and long short-term memory network are introduced for comparative experiments.The prediction result shows that the performance of photovoltaic(PV)power generation dynamic correction prediction based on integrated learning model fusion is significantly prior to other prediction models which introduced as comparison models,and the mean absolute error is improved by 10%~19%,which is also an advantage over the en-semble learning model that only uses weight fusion,with an increase of 14%.

关 键 词：光伏发电 功率预测 集成学习 权值融合 动态修正预测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]