摘要
光伏发电具有随机性、间歇性和波动性的特点,大规模光伏电站并网会严重影响电网的运行稳定性。提出了一种联邦学习架构下基于LSTM神经网络的光伏发电量预测算法解决方案。通过将历史太阳辐照度数据的统计特征与城市的天气预报类型相结合,创建了一种合成的天气预报。根据每天的时间和气象特征,为每个小时定义了独特的天气类型,对应不同的辐照度水平。合成天气预报融合了历史天气数据的统计特征,显著提高了预测准确性。研究结果显示,提出的光伏发电量预测算法可以提高预测的准确度。
Photovoltaic power generation is characterized by randomness,intermittency,and volatility,and the grid integration of large-scale photovoltaic power plants can significantly undermine the operational stability of the power grid.A solution based on an LSTM neural network for photovoltaic power generation forecasting under a federated learning architecture has been proposed.By integrating the statistical features of historical solar irradiance data with urban weather forecast types,a synthesized weather forecast has been created.Based on the time of day and meteorological characteristics,unique weather types corresponding to different irradiance levels have been defined for each hour.The synthesized weather forecast,which incorporates the statistical features of historical weather data,has significantly enhanced forecasting accuracy.The research results indicate that the proposed photovoltaic power generation forecasting algorithm can improve forecasting accuracy.
作者
韩忠修
王云鹏
HAN Zhongxiu;WANG Yunpeng(NARI Technology Co.,Ltd.,Nanjing,Jiangsu 211106,China)
出处
《自动化应用》
2025年第15期153-157,共5页
Automation Application
关键词
光伏发电
LSTM神经网络
电量预测
联邦学习架构
photovoltaic power
LSTM neural network
power prediction
federated learning architecture
