基于相似日与ISC-BiLSTM的短期光伏功率预测方法  

SHORT-TERM PHOTOVOLTAIC POWER FORECAST METHOD BASED ON SIMILAR DAYS AND ISC-BiLSTM

作　　者：杨轶航 韩璐 史华勃 邓鑫隆 陈梓桐 孙如田[3] Yang Yihang;Han Lu;Shi Huabo;Deng Xinlong;Chen Zitong;Sun Rutian(College of Electrical Engineering and Automation,Southwest Petroleum University,Chengdu 610500,China;State Grid Sichuan Electric Power Research Institute,Chengdu 610041,China;Karamay Vocational&Technical College,Karamay 834000,China)

机构地区：[1]西南石油大学电气信息学院,成都610500 [2]国网四川省电力公司电力科学研究院,成都610041 [3]克拉玛依职业技术学院,克拉玛依834000

出　　处：《太阳能学报》2025年第1期676-685,共10页Acta Energiae Solaris Sinica

基　　金：国家自然科学基金(51607151);电力物联网四川省重点实验室开放重点课题(PIT-F-202301)。

摘　　要：针对传统光伏功率预测方法的精度和鲁棒性难以兼顾的不足,提出一种结合相似日理论、改进麻雀算法(ISSA)与SE通道注意力机制的卷积(CNN)双向长短期记忆(BiLSTM)神经网络模型(简写为ISC-BiLSTM),能实现短期光伏功率的准确预测。该方法首先通过相关性计算,筛选出影响光伏功率的主要气象因子;再使用模糊C均值聚类(FCM)方法对存在相似天气特征的相似日进行聚类;然后通过加入SE的CNN对主要气象参数与历史功率的时空特征进行充分提取;接着利用BiLSTM对数据序列间的依赖关系进行捕捉;最后通过ISSA对模型的超参数进行寻优,并选择超参数最优的模型进行功率预测。对比实验与仿真结果表明,该方法预测误差较低,能实现日前分钟级短期光伏功率的准确预测。Aiming at the problem that the accuracy and robustness of traditional photovoltaic power prediction methods are difficult to balance,an improved convolutional(CNN)bidirectional long-term and short-term memory(BiLSTM)neural network model(ISC-BiLSTM)is proposed by combining similar days theory,improved sparrow algorithm(ISSA)and(SE)channel attention mechanism,which can achieve accurate prediction of short-term photovoltaic power.Firstly,the main meteorological factors affecting photovoltaic power are screened out by correlation calculation.Then the fuzzy C-means clustering(FCM)method is used to cluster similar days with similar weather characteristics.Then,the spatial and temporal characteristics of the main meteorological parameters and historical power are fully extracted by SE-CNN;then,BiLSTM is used to capture the dependencies between data sequences.Finally,the hyper-parameters of the model are optimized by ISSA,and the model with the optimal hyper-parameters is selected for power prediction.The comparative experiment and simulation results show that the prediction error of this method is low,and the accurate prediction of short-term photovoltaic power at the minute level before the day can be realized.

关 键 词：光伏发电 预测 神经网络 注意力机制 改进麻雀算法 模糊聚类 

分 类 号：TM615[电气工程—电力系统及自动化]