计及相似日的短期光伏功率预测  

The Short-Term Photovoltaic Power Forecast for Similar Days

作　　者：夏祥子 张春 XIA Xiangzi;ZHANG Chun(School of Electrical Engineering,Anhui Polytechnic University,Wuhu,Anhui 241000,China)

机构地区：[1]安徽工程大学电气工程学院,安徽芜湖241000

出　　处：《上海电力大学学报》2025年第2期112-119,共8页Journal of Shanghai University of Electric Power

基　　金：国家自然科学基金(U21A20146)。

摘　　要：为提高光伏电站输出功率的预测精度,提出了将长短期记忆(LSTM)模型与自适应增强器(Adaboost)相结合的光伏功率预测方法。首先,利用自组织神经映射聚类算法将输入的气象数据进行聚类,以提高预测的稳定性和一致性。然后,采用变分模态分解技术,将光伏功率序列拆分为具有不同频率的本征模态函数,从而有效降低光伏功率信号的复杂度和非平稳性。最后,利用改进麻雀优化算法(SSA)优化的LSTM模型训练多个弱预测器,将经过Adaboost修正的LSTM模型预测结果与原始LSTM模型预测结果相结合,以获得最终预测输出。仿真结果表明,该预测方法可有效提高日前高分辨率光伏功率预测精度和泛化能力。To enhance the accuracy of photovoltaic(PV)power output forecasting,a novel approach is proposed in this study.It integrates an improved sparrow search algorithm(SSA)for optimizing long short-term memory(LSTM)networks and combines LSTM with Adaboost.Initially,the meteorological input data is clustered using self-organizing map to enhance prediction stability and consistency.Subsequently,variational mode decomposition is employed to decompose the PV power sequence into intrinsic mode functions with varying frequencies,effectively reducing signal complexity and non-stationarity.Finally,multiple weak predictors are trained using the LSTM model optimized by the improved SSA.The predictions from the LSTM model,corrected by the Adaboost algorithm,are then combined with the original LSTM predictions to generate the final forecast output.Simulation results demonstrate that this combined forecasting model significantly improves the accuracy and generalization ability of high-resolution PV power forecasting for the next day.

关 键 词：光伏功率预测 神经网络 相似度分析 自适应增强器 自组织神经映射 

分 类 号：TM615[电气工程—电力系统及自动化]