基于GRO-SSA-LSTM的短期光伏发电功率预测  

SHORT TERM PHOTOVOLTAIC POWER PREDICTION BASED ON GRO-SSA-LSTM

作　　者：王玲芝 李晨阳 刘婧 李程 Wang Lingzhi;Li Chenyang;Liu Jing;Li Cheng(School of Automation,Xi’an University of Posts and Telecommunications,Xi’an 710121,China;State Grid Shaanxi Electric Power Co.,Ltd.,Weinan Power Supply Company,Weinan 714000,China)

机构地区：[1]西安邮电大学自动化学院,西安710121 [2]国网陕西省电力有限公司渭南供电公司,渭南714000

出　　处：《太阳能学报》2025年第2期401-409,共9页Acta Energiae Solaris Sinica

基　　金：国家自然科学基金(62073259;52177194)。

摘　　要：为提高光伏发电功率预测的精确度,保障电网能可持续稳定运行,将长短时记忆网络(LSTM)与淘金优化算法(GRO)改进后的麻雀搜索算法(SSA)结合起来,用于实现短期光伏发电功率的预测。首先,利用皮尔逊相关系数提取影响光伏功率的关键因素;然后,利用麻雀搜索算法对长短时记忆网络进行优化,得到网络中最优的隐含层节点数量、训练次数、学习率等超参数;其次,引入Tent混沌映射优化麻雀种群的初始分布,使得种群初始位置分布更加均匀;最后,为避免算法陷入局部最优,引入GRO对SSA进行优化,使得麻雀种群搜索范围更加广泛,结果更加精确。实验结果表明,与LSTM、SSA-LSTM相比,GROSSA-LSTM在短期光伏发电功率预测中具有更高的预测精度,且具有至关重要的现实意义。In order to improve the accuracy of photovoltaic power generation prediction and ensure the sustainable and stable operation of the power grid,this article combines long short term memory(LSTM)network with sparrow search algorithm(SSA)improved by gold rush optimizer(GRO)algorithm to achieve short-term photovoltaic power generation prediction.Firstly,the Pearson correlation coefficient is used to extract key factors that affect photovoltaic power;Secondly,the sparrow search algorithm is used to optimize the long and short term memory network,obtaining the optimal hyperparameters such as the number of hidden layer nodes,training times,and learning rate in the network.Then,introducing Tent chaotic mapping to optimize the initial distribution of sparrow population,making the initial position distribution of the population more uniform.Finally,to avoid the algorithm falling into local optima,GRO is introduced to optimize SSA,making the search range of sparrow population more extensive and the results more accurate.The experimental results indicate that compared with LSTM and SSA-LSTM,GRO-SSA-LSTM has higher prediction accuracy in short-term photovoltaic power generation prediction and has crucial practical significance.

关 键 词：光伏发电 预测模型 长短时记忆网络 麻雀搜索算法 淘金优化算法 

分 类 号：TM615[电气工程—电力系统及自动化]