基于集成聚类和XGBoost的短期光伏发电功率预测Short-term Photovoltaic Power Prediction Based on Ensemble Clustering and XGBoost
常俊晓,金之榆,卢姬,吴思圆
CHANG Junxiao,JIN Zhiyu,LU Ji,WU Siyuan
摘要(Abstract)：
光伏发电功率预测是光伏发电规划和电网经济运行的重要基础。针对K-means算法无法确定最佳聚类数和聚类结果不稳定的问题,以及GBDT(梯度提升树)预测结构简单、预测精度低的缺点,提出集成聚类和XGBoost的组合预测算法用于构建短期光伏发电功率的预测模型。首先,采用Meanshift(均值漂移算法)、凝聚层次聚类算法和DBI(戴维森堡丁指数)优化K-means算法,解决了无法确定最佳聚类数和聚类结果不稳定的问题;其次,在集成聚类得到的聚类结果数据上,训练得到不同气象类型下的XGBoost预测模型,以此来对光伏发电功率进行短期预测;最后,为了验证XGBoost集成学习算法的有效性,采用GBDT算法对比预测结果。经分析验证,所提出的组合预测算法在晴天、多云、阴雨天的气象类型下具有更高的预测精度,验证了该方法的有效性。
Photovoltaic power prediction is an important basis for photovoltaic power generation planning and economic operation of power grid. As the K-means algorithm cannot determine the optimal number of clusters with unstable clustering results, and Gradient Boosting Decision Tree(GBDT) is simple in prediction structure with low prediction accuracy, a combined prediction algorithm integrating clustering and XGBoost is proposed to build a short-term photovoltaic power prediction model. Firstly, Mean-shift algorithm, agglomerative hierarchical clustering algorithm and Davies-Bouldin Index Optimization K-means algorithm are used to solve the problem that the optimal clustering number cannot be determined and the clustering result is unstable.Secondly, based on the clustering result data obtained by ensemble clustering, the XGBoost prediction model under different weather types is trained to predict the short-term photovoltaic power generation. Finally, GBDT algorithm is employed to compare the forecasting result to prove the effectiveness of the proposed method.The analysis proves that the proposed combination prediction algorithm can get more accurate prediction results on sunny, cloudy and rainy days, which verifies the effectiveness of the method.
关键词(KeyWords)：
集成聚类;XGBoost算法;光伏发电;短期功率预测
ensemble clustering;XGBoost algorithm;photovoltaic power generation;short-term power prediction
基金项目(Foundation): 国网浙江省电力有限公司科技项目（5211TZ1900S7）
作者(Author):
常俊晓,金之榆,卢姬,吴思圆
CHANG Junxiao,JIN Zhiyu,LU Ji,WU Siyuan
DOI: 10.19585/j.zjdl.202110013
参考文献(References)：
- [1]荆博，谭伦农，钱政，等.光伏发电短期预测研究进展综述[J].电测与仪表，2017,54(12):1-6.
- [2]邓呈格，钟志峰.人工智能算法在光伏发电量预测中的应用[J].物联网技术，2017,7(4):68-69.
- [3]DIAGNE M,DAVID M,LAURET P,et al.Review of solar irradiance forecasting methods and a proposition for smallscale insular grids[J].Renewable and Sustainable Energy Reviews,2013,27:65-76.
- [4]张华彬，杨明玉.基于天气类型聚类和LS-SVM的光伏出力预测[J].电力科学与工程，2014,30(10):42-47.
- [5]JIMNEZ-PREZ P F, MORA-LóPEZ L.Modeling and forecasting hourly global solar radiation using clustering and classification techniques[J].Solar Energy,2016,135:682-691.
- [6]赵唯嘉，张宁，康重庆，等.光伏发电出力的条件预测误差概率分布估计方法[J].电力系统自动化，2015,39(16):8-15.
- [7]陈瑶，陈晓宁.基于自适应K-means和LSTM的短期光伏发电预测[J/OL].电测与仪表：1-5[2021-04-17].http://kns.cnki.net/kcms/detail/23.1202.TH.20200521.1438.014.html.
- [8]邵堃侠，郭卫民，杨宁，等.基于K-means算法的RBF神经网络预测光伏电站短期出力[J].上海电机学院学报，2017,20(1):27-33.
- [9]CHEN T Q,GUESTRIN C.XGBoost:a scalable tree boosting system[C]//Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York,NY,USA:ACM,2016.
- [10]彭曙蓉，郑国栋，黄士峻，等.基于XGBoost算法融合多特征短期光伏发电量预测[J].电测与仪表，2020,57(24):76-83.
- [11]唐雅洁，林达，倪筹帷，等.基于XGBoost的双层协同实时校正超短期光伏预测[J].电力系统自动化，2021,45(7):18-27.
- [12]SHEN S,LIU Q Y,TAO X C,et al.Application of the XGBOOST on the assessment of transient stability of power system[C]//Proceedings of the 2019 International Conference on Electronical,Mechanical and Materials Engineering(ICE2ME 2019).Paris,France:Atlantis Press,2019.
- [13]陈中，宗鹏鹏.基于样本扩张灰色关联分析的光伏出力预测[J].太阳能学报，2017,38(11):2909-2915.
- [14]CAPóM,PREZ A,LOZANO J A.An efficient approximation to the K-means clustering for massive data[J].Knowledge-Based Systems,2017,117:56-69.