基于数据平稳化和BiLSTM的短期风电功率预测方法  

Short-term wind power prediction method based on data stabilization and BiLSTM

作　　者：唐贤伦[1] 张家瑞 郭祥麟 邹密 TANG Xianlun;ZHANG Jiarui;GUO Xianglin;ZOU Mi(School of Automation,Chongqing University of Posts and Telecommunications,Chongqing 400065,P.R.China)

机构地区：[1]重庆邮电大学自动化学院,重庆400065

出　　处：《重庆邮电大学学报（自然科学版）》2023年第6期1135-1144,共10页Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)

基　　金：国家自然科学基金项目(60905066);重庆市教委科技项目(KJZD-M202200603,KJZD-K202202401)。

摘　　要：针对风电功率数据序列波动大、随机性强、非线性以及选取输入变量困难的问题,提出一种结合自适应噪声完备集合经验模态分解(complete ensemble empirical mode decomposition with adaptive noise,CEEMDAN)和双向长短期记忆网络(bidirectional long short term memory,BiLSTM)的短期风电功率预测组合模型。通过CEEMDAN对原始功率数据序列进行分解及平稳化处理,并根据各分量样本熵值进行合并重构;利用偏自相关函数(partial autocorrelation function,PACF)计算各重构分量的滞后期数,以此确定各重构分量在BiLSTM网络模型中的最佳输入变量;根据各重构分量的预测值相加得到最终预测结果。实验结果表明,与几种传统的单一预测模型和组合预测模型相比,提出的模型具有更优的预测结果和更高的预测精度。Aiming at the characteristics of large fluctuation,strong randomness and nonlinearity of wind power data series,and the problem of how to select input variables ignored in most current wind power prediction studies,we propose a short-term wind power prediction model combining complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)and bidirectional long short-term memory(BiLSTM).Firstly,the original power data sequence is decomposed and stabilized by CEEMDAN,and then merged and reconstructed according to the sample entropy value of each component.Then,the lag period of each reconstructed component is calculated by partial autocorrelation function(PACF)to determine the optimal input variable of each reconstructed component in the BiLSTM network model.Lastly,the final prediction result is obtained by adding the predicted values of the reconstructed components.Experimental results of two examples show that,compared with several traditional single prediction models and combined prediction models,the proposed method has better prediction results and more precise prediction accuracy.

关 键 词：短期风电功率预测 自适应噪声完备集合经验模态分解 双向长短期记忆网络 偏自相关函数 

分 类 号：TN86[电子电信—信息与通信工程] TM614[电气工程—电力系统及自动化]