摘要
光伏出力的有效预测估计数据对电网的优化运行与规划设计具有重要意义。针对传统回归模型难以有效提取光伏数据的规律信息和深度学习模型容易导致过拟合的问题,提出一种基于ResNet-GRU的深度学习模型。首先,对大量历史数据进行清洗,剔除异常值,进行时间编码,根据相关性高低选取相应的气象元素与历史功率作为预测模型的输入特征;然后,采用残差卷积神经网络对输入进行特征提取,输出到门循环单元对历史功率进行时序拟合,输出光伏发电功率的预测值;最后,以某地区光伏系统数据为仿真算例进行分析,结果表明,所提出的模型与传统深度学习模型相比可以有效提高预测精度,验证了该模型的有效性。
The effective prediction data of photovoltaic output is of great significance to the optimal operation and planning design of power grid.Aiming at the problem that traditional regression model is difficult to effectively extract the rule information of photovoltaic data and deep learning model is easy to lead to overfitting,this paper proposes a deep learning model based on ResNet-GRU.Firstly,a large number of historical data were cleaned,outliers were removed,time coding was carried out,and several meteorological factors with high correlation with PV power and historical power deviation data were selected as input features of the prediction model.Secondly,the residual convolutional neural network(ResNet)was used to extract the feature of the input,and the output was sent to the gate cycle unit(GRU)for the time sequence fitting of the historical power,and the predicted value of the photovoltaic power was output.Finally,taking the photovoltaic system data in Shandong province,China as an simulation example for analysis,the results show that compared with the traditional deep learning model,the proposed model can effectively improve the prediction accuracy and verify the effectiveness of the proposed model.
作者
孙毅
刘蕊
魏孟迪
王宪
孙东磊
SUN Yi;LIU Rui;WEI Mengdi;WANG Xian;SUN Donglei(Economic&Technological Research Institute,State Grid Shandong Electric Power Company,Jinan 250021,China;School of Electrical Automation and Information Engineering of Tianjin University,Tianjin 300072,China)
出处
《供用电》
2023年第6期34-41,共8页
Distribution & Utilization
基金
国网山东省电力公司科技项目“考虑开发潜力与系统约束的山东新能源电力时空布局研究”(52062522000N)。
关键词
光伏功率预测
残差神经网络
门控循环单元
时间编码
功率偏差预测
photovoltaic power prediction
residual neural network
gate recurrent unit
time encoding
power deviation prediction
