基于优化的VMD-CNN-LSTM模型的光伏功率预测  

Photovoltaic power prediction based on optimized VMD-CNN-LSTM model

作　　者：吴勇 高昕[1] 郭灏阳 刁海岸 刘庆丰 杨强强 WU Yong;GAO Xin;GUO Haoyang;DIAO Hai’an;LIU Qingfeng;YANG Qiangqiang(School of Electrical and Information Engineering,Anhui University of Science and Technology,Huainan 232001,China)

机构地区：[1]安徽理工大学电气与信息工程学院,安徽淮南232001

出　　处：《邵阳学院学报（自然科学版）》2022年第6期9-17,共9页Journal of Shaoyang University：Natural Science Edition

基　　金：2018安徽省自然科学基金(1808085MF169);安徽理工大学研究生创新基金项目(2021CX2065)。

摘　　要：针对变分模态分解(variational mode decomposition,VMD)用经验知识定义模态数、传统功率预测方法缺少对时序数据考虑等问题,提出一种基于优化VMD、联合卷积神经网络(convolutional neural network,CNN)-长短期记忆(long short-term-memory,LSTM)网络的组合预测模型。精度更高的光伏预测可以提高光伏并网的安全性、可靠性。通过布谷鸟搜索(cuckoo search,CS)算法优化VMD的主要参数,把光伏功率分解成若干趋于稳定的模态分量;随后将分解量送入联合CNN与LSTM的组合预测模型进行逐一预测并将功率预测值进行叠加评估,建立起基于CS-VMD-CNN-LSTM的光伏组合预测模型。以宁夏太阳山光伏电站的实测数据做仿真分析,结果表明,较遗传算法(genetic algorithm,GA)优化VMD等模型,该模型对预测光伏功率更具有效性,预测结果更优。Aiming at the problems that variational mode decomposition(VMD)used empirical knowledge to define mode number and traditional power prediction methods lack consideration of time series data,a new method based on optimized VMD was proposed.The combined prediction model of convolutional neural network(CNN)and long short-term memory(LSTM)network can improve the security and reliability of photovoltaic grid-connected.The main parameters of VMD were optimized by cuckoo search(CS)algorithm,and the photovoltaic power was decomposed into several stable modal components;then,the decomposition quantity was sent to the combined prediction model of CNN and LSTM to predict one by one.The predicted power value is evaluated superimposed to establish the combined prediction model of PV based on CS-VMD-CNN-LSTM.Based on the measured data of Ningxia Sunshan photovoltaic power station,the simulation results show that,compared with the genetic algorithm(GA)to optimize VMD and other models,the model is more effective in predicting the photovoltaic power,and the prediction results are better.

关 键 词：功率预测 布谷鸟优化 变分模态分解 卷积神经网络 长经期记忆网络 

分 类 号：TM615[电气工程—电力系统及自动化]