基于EEMD-Kmeans...LSTM的短期光伏功率预测_朱坤.pdf

1、2023.1Vol.47No.1研 究 与 设 计收稿日期：2022-06-29基金项目：国家自然科学基金(61871410)；珠海市产学研项目(ZH22017001200053PWC)；佛山市深入推进创新驱动助力工程项目(2021003)作者简介：朱坤(1996)，男，四川省人，硕士，主要研究方向为太阳能光伏发电技术。通信作者：付青，E-mail：基于EEMD-Kmeans-ALO-LSTM的短期光伏功率预测朱 坤，付 青(中山大学 物理学院，广东 广州 510275)摘要：光伏功率预测对电网调度具有重要意义。针对光伏功率数据具有较强波动性和不稳定性的特点，提出了一种基 于 集 成 经 验

2、模 态 分 解(ensemble empirical mode decomposition,EEMD)、K 均 值 聚 类 算 法(Kmeans clusteringalgorithm,Kmeans)和蚁狮优化(ant lion optimization,ALO)算法优化的长短期记忆神经网络(long short-term memorynetwork,LSTM)的光伏功率组合预测模型。对光伏功率数据进行 EEMD 分解，得到相应的本征模态分量(intrinsicmode function，IMF)和残差项；引入 Kmeans聚类对分解后的序列重构，降低序列复杂度和分量数量；将重构后的子序列输入

3、经 ALO优化的 LSTM 模型进行预测，并将各序列预测结果简单加和作为最终预测值。与目前应用较广泛的EEMD-LSTM模型对比，表明EEMD-Kmeans-LSTM和EEMD-Kmeans-ALO-LSTM模型的预测精度均得到一定程度的提高。关键词：Kmeans聚类；集成经验模态分解；蚁狮优化算法；长短期记忆神经网络；光伏功率预测中图分类号：TM 615文献标识码：A文章编号：1002-087 X(2023)01-0103-05DOI:10.3969/j.issn.1002-087X.2023.01.023A photovoltaic power forecasting method bas

4、ed on EEMD-Kmeans-ALO-LSTMZHU Kun,FU Qing(School of Physics,Sun Yat-sen University,Guangzhou Guangdong 510275,China)Abstract:Photovoltaic power prediction has significance for power grid dispatching.This article focuses on thecharacteristics of volatility and instability in photovoltaic power,propos

5、es a combination forecasting model using thelong short-term Memory(LSTM)network optimized by ant lion optimization(ALO)algorithm based on ensembleempirical mode decomposition(EEMD)and Kmeans clustering algorithm(Kmeans).First,the photovoltaic powerdata is decomposed by EEMD.The corresponding intrins

6、ic mode functions and residual are obtained.Then,thedecomposed sequence is reconstructed by Kmeans clustering to reduce the sequence complexity and the number ofdecomposed components.Finally,the reconstructed subsequence is input into the LSTM model optimized by ALO forprediction,and the prediction results of each sequence are simply summed as the final prediction value.Comparedwith the currently widely used EEMD-LSTM algorithm,the prediction accuracy of the EEMD-Kmeans-LST