基于数据清洗与组合学习的光伏发电功率预测方法研究  

Short-term photovoltaic forecasting based on data cleansing and model combination

作　　者：邱明 鲁冠军[2] 吴昊天[3] 杨仲卿 Qiu Ming;Lu Guanjun;Wu Haotian;Yang Zhongqing(School of Energy and Power Engineering,Chongqing University,Chongqing 400044,China;School of Power Engineering,Chongqing Electric Power College,Chongqing 400053,China;School of Electrical and Electronic Engineering,North China Electric Power University,Beijing 102206,China)

机构地区：[1]重庆大学能源与动力工程学院,重庆400044 [2]重庆电力高等专科学校动力工程学院,重庆400053 [3]华北电力大学电气与电子工程学院,北京102206

出　　处：《可再生能源》2020年第12期1583-1589,共7页Renewable Energy Resources

基　　金：中央高校基本科研业务费专项资金(2019XS29);国家自然科学基金(59577060)。

摘　　要：短期光伏发电功率预测对电网调度计划的安排和优化具有重要意义。机器学习类算法的飞速进步有利于提高光伏发电功率的预测精度。文章提出了一种基于数据清洗与组合学习的光伏发电功率预测方法。考虑到变电站实际运行时,在通信与传输过程中会出现数据遗漏的状况,在数据输入模型前,采用KNN算法对缺失数据进行补全;然后,将极限学习机、Adaboost模型和神经网络模型的预测结果进行动态组合,并通过Lasso算法在一定周期内对权值进行更新,获得最终预测结果;最后,利用北京大兴区的实际光伏发电数据来验证文章所提出的预测算法的准确性。模拟结果表明,在晴天和阴雨天条件下,组合学习模型预测结果均比较准确。Short-term photovoltaic(PV)generation forecasting has very important influence for the dispatching and optimal operation of power system.The machine learning and artificial intelligence technology has made great contribution on the accuracy development.A short-term photovoltaic power forecasting method based on data cleansing and model combination is proposed in this paper.Considering that data mission condition in the process of transmission and communication,the KNN algorithm is used to complement the missing section,then extreme learning machine,Adaboost and artificial neural network are dynamic combined by the Lasso algorithm,whose weight can be updated periodically to obtain the final forecasting result.At last,the actual data of Beijing photovoltaic power generation system is used to verify the accuracy of the prediction results.

关 键 词：短期光伏发电功率预测 数据清洗 组合学习 Lasso 极限学习机 ADABOOST 

分 类 号：TK51[动力工程及工程热物理—热能工程] TM715[电气工程—电力系统及自动化]