基于改进鱼鹰优化算法与VMD-LSTM的超短期风电功率预测  

ULTRA SHORT-TERM WIND POWER PREDICTION BASED ON IMPROVED OSPERY OPTIMIZATION ALGORITHM AND VMD-LSTM

作　　者：罗潇远 刘杰 杨斌 覃涛[1] 陈昌盛[1] 杨靖[1,3] Luo Xiaoyuan;Liu Jie;Yang Bin;Qin Tao;Chen Changsheng;Yang Jing(Electrical Engineering College,Guizhou University,Guiyang 550025,China;China Power Construction Group Guizhou Engineering Co.,Ltd.,Guiyang 550025,China;Guizhou Provincial Key Laboratory of Internet+Intelligent Manufacturing,Guiyang 550025,China)

机构地区：[1]贵州大学电气工程学院,贵阳550025 [2]中国电建集团贵州工程有限公司,贵阳550025 [3]贵州省互联网+协同智能制造重点实验室,贵阳550025

出　　处：《太阳能学报》2025年第3期652-660,共9页Acta Energiae Solaris Sinica

基　　金：贵州省科技支撑计划(黔科合支撑[2023]一般411;黔科合支撑[2023]一般412;黔科合支撑[2024]一般051);贵州省双碳研究院开放课题(DCRE-2023-13)。

摘　　要：为提升超短期风电功率的预测精度,提出一种加入融合柯西变异和反向学习策略的改进鱼鹰优化算法(IOOA),用于优化以长短期记忆网络(LSTM)和变模态分解(VMD)为基础的组合预测模型。首先,采用变模态分解收集的历史风电功率数据,将非线性较强的原始功率数据分解为较为稳定的子序列。其次,使用改进鱼鹰优化算法对长短期记忆网络的隐藏单元数目、训练周期、初始学习率3个参数进行寻优。最后,使用长短期记忆网络对各子序列预测,将各子序列预测值叠加起来得到最终结果。通过风电场实测数据仿真分析,相比于普通长短期记忆网络模型的预测结果,所提模型的均方根误差下降了62.5%、平均绝对百分比误差和平均绝对误差分别下降了61.1%和55.9%,预测精度也高于其他4种组合预测模型,表明该模型成功提高了超短期风电功率的预测精度。To enhance the accuracy of ultra-short-term wind power,a new optimization algorithm called improved osprey optimization algorithm(IOOA)is proposed.It combines Cauchy mutation and reverse learning strategy to optimize a combination prediction model based on long short-term memory network(LSTM)and variable mode decomposition(VMD).Firstly,the historical wind power data collected through VMD is used to decompose the original power data with strong nonlinearity into relatively stable subsequences.Secondly,the osprey optimization algorithm,combining Cauchy mutation and reverse learning strategies,is used to optimize the number of hidden units,training period,and initial learning rate of the LSTM.Finally,the final prediction result is obtained by using a LSTM to predict each subsequence,and the predicted values of each subsequence were superimposed to obtain the final result.The proposed wind farm prediction model has been analyzed by using measured data.The results were compared with ordinary short-term memory neural network models.The proposed model reduced the RMSE by 62.5%,decreased MAPE by 62.2%and MAE by 55.9%.And the prediction accuracy is also higher than other four combination prediction models,indicating its success in improving the prediction accuracy of short-term wind power.

关 键 词：长短期记忆网络 变模态分解 风力发电 改进鱼鹰优化算法 功率预测 优化算法 

分 类 号：TM614[电气工程—电力系统及自动化]