考虑VMD残差量和优化BiLSTM的短期负荷预测  

Short-term Load Forecasting Considering VMD Residuals and Optimizing BiLSTM

作　　者：谢煜轩 王红君[1] 岳有军[1] 赵辉[1] XIE Yuxuan;WANG Hongjun;YUE Youjun;ZHAO Hui(School of Automation Tianjin Complex Control Theory and Application of Key Laboratory,Tianjin University of Technology,Tianjin 300384,China)

机构地区：[1]天津理工大学天津市复杂控制理论与应用重点实验室,天津300384

出　　处：《复杂系统与复杂性科学》2024年第4期149-156,共8页Complex Systems and Complexity Science

基　　金：天津市自然科学基金重点项目(08JCZDJC18600);天津市教委重点基金项目(2006ZD32)。

摘　　要：为进一步提高短期负荷预测精度,提出了一种基于变分模态分解(VMD)并考虑VMD残差量和改进北方苍鹰算法(INGO)优化双向长短时记忆(BiLSTM)网络的短期负荷预测方法。首先利用VMD将历史负荷数据分解为多个本征模分量(IMFs)和一个残差量。再将各IMF和残差量以及相关气象参数分别构建BiLSTM模型进行预测。为避免因超参数选取不佳对预测精度的影响,采用INGO对BiLSTM的隐含层节点、训练次数、学习率进行优化。最后将预测结果叠加得出最终结果。通过具体算例分析,将本文采用方法与其他方法对比,具有较高的预测精度,验证了本文方法的有效性。This study proposes a new method to improve short-term load forecasting accuracy.The method is based on Variational Modal Decomposition(VMD)with consideration of VMD residuals and an Improved Northern Eagle Algorithm(INGO)optimized Bi-directional Long Short Term Memory(BiLSTM)network.The VMD is used to decompose historical load data into multiple eigenmode components(IMFs)and a residual quantity.The BiLSTM model is then constructed separately for each IMF and residual,as well as the associated meteorological parameters.To avoid the impact of poorly selected hyperparameters on prediction accuracy,the INGO algorithm optimizes the implied layer nodes,training times,and learning rates of the BiLSTM.Last but not least,the prediction results are superimposed to obtain the final results.By analyzing specific cases,this paper′s method has demonstrated a higher prediction precision when compared to alternative methods.This validation confirms the effectiveness of the method presented in this article.

关 键 词：短期负荷预测 变分模态分解 北方苍鹰算法 双向长短时记忆网络 

分 类 号：TM715[电气工程—电力系统及自动化] O224[理学—运筹学与控制论]