EO-REEVMD-BILSTM的两阶段超短期风电功率预测
Prediction of a two-stage ultra-short-term wind power based on EO-REEVMD-BILSTM
赵为光 1梁桐 1杨莹 1刘振羽 1曹美萱 1徐欢欢1
作者信息
- 1. 黑龙江科技大学 电气与控制工程学院,哈尔滨 150022
- 折叠
摘要
为提升风电功率的预测精度,提出一种基于平衡优化器优化算法的两阶段超短期风电功率预测方法.通过建立一种适用于风电功率预测的回归包络熵适应度目标模型,利用平衡优化器算法寻优VMD分解参数,实现对原始风电功率信号的合理分解,有效减小分解损失.基于BILSTM神经网络模型分别预测分解的模态分量,根据叠加各分量的预测结果获得初步风电预测功率序列,利用误差预测值纠正上一阶段预测结果.以土耳其某地区提供的风电功率数据作为实际算例,通过仿真实验与实测风电功率比较.结果表明:文中所提方法的 RMSE 与 MAE 仅为28.378 1与17.429 7,R2 为0.998 6,明显低于BILSTM等单一预测模型与其他组合预测方法,验证了文中方法的有效性.
Abstract
This paper seeks to improve the prediction accuracy of wind powe,and proposes a two-stage ultra-short-term wind power prediction method based on equilibrium optimizer optimization algo-rithm.The study involves establishing a regression envelope entropy fitness target model suitable for wind power prediction to optimize the VMD decomposition parameter by using the equilibrium optimizer algorith for a reasonable decomposition of the original wind power signal and for the effective decrease of the de-composition loss;separately predicting the modal components decomposed based on the BILSTM neural network model;obtaining the preliminary wind power prediction sequence according to the prediction re-sults of the superimposed components;and correcting the prediction results of the previous stage.By com-paring the simulation experiments with the measured wind power and using the wind power data provided,the experimental results show that,by taking a region in Turkey as a practical example,RMSE and MAE in the proposed method are only 28.378 1 and 17.429 7,and R2 is 0.998 6,which is significantly lower than that of the single prediction model such as the BILSTM and the other combination of prediction meth-ods,verifying the validity of the method in this paper.
关键词
风电功率/平衡优化器/变分模态分解/BILSTM/误差纠正Key words
wind power/equilibrium optimizer algorithm/variational modal decomposition/bidirec-tional long and short term memory neural network/error correction引用本文复制引用
基金项目
黑龙江省省属高校基本科研业务费项目(2019-KYYWF-0731)
出版年
2024
国家科技期刊平台
NSTL
维普
万方数据