考虑相似日相关信息的基于FCM-UW-ADAGRU的超短期光伏预测方法  

FCM-UW-ADAGRU-based Ultra-short-term PV Prediction Considering Relevant Information of Similar Days

作　　者：施静辉 高翔 王瑞林 SHI Jinghui;GAO Xiang;WANG Ruilin(Envision Energy Co.,Ltd.,Wuxi 214441,China;Shenzhen Polytechnic University,Shenzhen 518055,China;College of Electric Engineering,Shanghai University of Electric Power,Shanghai 200090,China)

机构地区：[1]远景能源有限公司,江苏无锡214441 [2]深圳职业技术学院,广东深圳518055 [3]上海电力大学电气工程学院,上海200090

出　　处：《电工技术》2024年第16期58-63,共6页Electric Engineering

基　　金：深圳市高层次人才科研创业基金(编号6022310042k)。

摘　　要：提出了一种模糊C均值聚类和不确定性加权自适应门控单元神经网络(FCM-UW-ADAGRU)模型对日前分钟级光伏出力进行预测。首先,基于FCM对历史日天气进行划分,采用历史功率数据的5个统计指标(协调平均值、几何平均值、变异系数、峰度和偏度)作为聚类特征。其次,通过分布识别模块从相同天气类型的相似日样本中识别出不同的数据分布,并通过分布匹配模块从所有相似日数据中挖掘相关信息,以处理未来可能遇到的未知气象信息。最后,基于不确定性加权(UW)平衡预测误差和相关信息误差,提高模型训练精度。与现有方法的比较实验表明该方法具有较高的精度和鲁棒性,验证了模型的有效性。The present work studied and proposed a fuzzy C-means clustering and uncertainty-weighted adaptive gating unit neural network(FCM-UW-ADAGRU)model,aiming at the prediction of day-ahead minute-level PV output.First the historical daily weather was classified through FCM by using 5 statistical indexes of historical power data,i.e.coordinated mean,geometric mean,coefficient of variation,kurtosis,and skewness,as the clustering characteristics.Then different data distributions from similar day samples of the same weather type were identified by distribution recognition module,and the relevant information can be mined from all similar daily data through the distribution matching module,in order to deal with possible unknown meteorological information about the future.Finally balance prediction error and related information error based on uncertainty weighting(UW)were employed to improve model training accuracy.In a case study,the above method was compared to currently prevailing methods,demonstrating its higher accuracy and robustness,and verifying the effectiveness of the proposed model.

关 键 词：自适应门控单元神经网络 模糊C-均值聚类 光伏超短期预测 相似日 

分 类 号：TM715[电气工程—电力系统及自动化]