电力系统及其自动化学报2025，Vol.37Issue(4)：78-87,10.DOI:10.19635/j.cnki.csu-epsa.001488
基于序列成分重组与时序自注意力机制改进TCN-BiLSTM的短期电力负荷预测
Short-term Power Load Forecasting Based on Sequence Component Recombination and Temporal Self-attention Mechanism Improved TCN-BiLSTM
摘要
Abstract
To address the issue of low accuracy in regional power load forecasting,a short-term load forecasting method based on sequence component recombination and temporal self-attention mechanism improved temporal convolutional network-bidirectional long short-term memory network(TCN-BiLSTM)is proposed in this paper.First,the optimal ini-tial decomposition number is determined through the central frequency method,and then the original load sequence is decomposed using the variational mode decomposition algorithm to obtain multiple sequences of different frequency components.Second,the multiple component sequences are clustered by K-means based on the sample entropy of each component sequence,thus obtaining the recombined load sequence components with the optimal number of clusters.Third,each recombined component is input into the load forecasting model proposed in this paper to obtain the predic-tion results of each recombined component.Finally,the prediction results of recombined components are linearly su-perimposed to obtain the final load prediction results.The analysis of a case study shows that compared with the average value of other related contrast models,the prediction root mean square error of the proposed method is reduced by 46.37％,and the model fitting effect is improved by 3.24％on average.This result indicates that the proposed method possesses a high prediction accuracy and a better model fitting effect,which is suitable for regional power load forecasting.关键词
负荷预测/变分模态分解/样本熵/K均值聚类/时序自注意力机制/时间卷积网络/双向长短期记忆网络Key words
load forecasting/variational mode decomposition/sample entropy/K-means clustering/temporal self-attention mechanism/temporal convolutional network(TCN)/bidirectional long short-term memory network(BiLSTM)分类
动力与电气工程引用本文复制引用
易雅雯,娄素华..基于序列成分重组与时序自注意力机制改进TCN-BiLSTM的短期电力负荷预测[J].电力系统及其自动化学报,2025,37(4):78-87,10.基金项目
国家电网公司科学技术项目(5108-202218280A-2-429-XG). （5108-202218280A-2-429-XG）
