基于混合深度学习的光伏集群发电预测  

Photovoltaic Cluster Power Generation Prediction Based on Hybrid Deep Learning

作　　者：罗兴 胡子健 马洲俊 吕湛 LUO Xing;HU Zijian;MA Zhoujun;LYU Zhan(Nanjing Power Supply Company,State Grid Jiangsu Electric Power Co.Ltd.,Nanjing 210024,China)

机构地区：[1]国网江苏省电力有限公司南京供电分公司,南京210024

出　　处：《西南大学学报(自然科学版)》2025年第3期200-210,共11页Journal of Southwest University(Natural Science Edition)

基　　金：国家重点研发计划项目(2022YFB2404200);国网江苏省电力有限公司科技项目(J2023017)。

摘　　要：光伏发电作为可再生能源在实现能源转型和减缓气候变化方面具有重要意义,然而光伏发电预测面临诸多挑战,包括天气不确定性、数据质量问题、复杂的影响因素以及模型复杂度和计算成本等,并且往往只考虑单个站点的相关因素,并没有考虑多个站点之间的相互影响。传统预测方法泛化能力有限,不能很好地捕捉复杂的非线性关系,缺乏足够的灵活性。为解决这些问题,提出基于混合深度学习的光伏集群发电预测框架,利用站点历史发电数据计算互信息并形成集群网络结构,然后构建具有不同特征的经典深度学习模型进行预测,包括长短期记忆网络(LSTM)、卷积神经网络(CNN)、时序卷积网络(TCN)以及极端梯度提升(XGBOOST),最后通过自适应权重计算实现模型集成学习并完成预测。对国家可再生能源实验室(NREL)真实光伏发电量数据进行预测,实验结果表明:集成模型具有更好的泛化能力并且在预测精度上相较单一模型也有显著提升。Photovoltaic(PV)power generation,as a renewable energy source,is considered significant in achieving energy transformation and mitigating the impacts of climate change.Despite its potential,the prediction of photovoltaic power generation faces several challenges,including weather uncertainty,data quality issues,complex influencing factors,model complexity and computational costs.Moreover,existing prediction methods often only consider relevant factors at a single site,neglecting the interactions between multiple sites.These traditional approaches have limited the generalization ability and struggled to capture the complex nonlinear relationships effectively,lacking the necessary flexibility.To address these issues,a hybrid deep learning-based photovoltaic cluster power generation prediction framework was proposed in this paper.Firstly,historical power generation data from stations were utilized to calculate mutual information for forming a cluster network structure.Subsequently,classic deep learning models with different features were constructed for prediction,including the Long Short-Term Memory Network(LSTM),Convolutional Neural Network(CNN),Time Series Convolutional Network(TCN),and Extreme Gradient Boosting(XGBOOST).Finally,model ensemble learning was achieved through adaptive weight calculation to complete the prediction.Real photovoltaic power generation data from the National Renewable Energy Laboratory(NREL)were utilized in this study for prediction.The experimental results demonstrate that the integrated model exhibits superior generalization ability and significantly improves the prediction accuracy compared to individual model.

关 键 词：混合深度学习 光伏集群发电预测 自适应权重 互信息 

分 类 号：TM615[电气工程—电力系统及自动化] TP183[自动化与计算机技术—控制理论与控制工程]