引用本文:祁宇轩,范俊岩,吴定会,汪晶.基于相似日与BiLSTM组合的短期电力负荷预测[J].控制理论与应用,2024,41(12):2304~2314.[点击复制]
QI Yu-xuan,FAN Jun-yan,WU Ding-hui,WANG Jing.Short term power load forecasting based on the combination of similar days and BiLSTM[J].Control Theory and Technology,2024,41(12):2304~2314.[点击复制]
基于相似日与BiLSTM组合的短期电力负荷预测
Short term power load forecasting based on the combination of similar days and BiLSTM
摘要点击 3588  全文点击 51  投稿时间：2022-11-02  修订日期：2024-08-21
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2023.20969
  2024,41(12):2304-2314
中文关键词  短期电力负荷预测  相似日  深度学习  鲸鱼优化算法  变分模态分解
英文关键词  short-term power load forecasting  similar day  deep learning  whale optimization algorithm  variational modal decomposition
基金项目  国家重点研发项目(2020YFB1711100, 2020YFB1711102)资助.
作者单位E-mail
祁宇轩 江南大学 nldqyx@163.com 
范俊岩 江南大学  
吴定会* 江南大学 wdh123@jiangnan.edu.cn 
汪晶 上海宝信软件有限公司  
中文摘要
      短期电力负荷存在非线性、波动性和影响因素多等特征, 针对以上特征所导致的预测精度不足, 本文提出一种基于相似日与双向长短时记忆神经网络(BiLSTM)组合的短期电力负荷预测模型. 首先, 剖析电力负荷的动态变化机理, 基于相似日和灰色关联分析方法, 构建负荷与特征融合数据集; 其次, 采用变分模态分解(VMD)方法将高波动、非线性的原始负荷数据分解为多个相对平稳的分量, 并对各分量分别搭建BiLSTM预测模型; 最后, 采用鲸鱼算法(WOA)对模型的分解参数和相似日天数进行优化, 减小模型的固有误差. 以新英格兰某地区的实际数据进行仿真验证, 所提模型的平均绝对百分比误差(MAPE)、平均绝对误差(MAE)和均方根误差(RMSE)分别为0.58%, 42,78, 均优于对照模型, 有效提升了负荷预测精度.
英文摘要
      Short term power load has the characteristics of nonlinearity, volatility and many influencing factors. Aiming at the lack of forecasting accuracy caused by the above characteristics, a short-term power load forecasting model based on the combination of similar days and bi directional long short memory neural network (BiLSTM) is proposed. First, the dynamic change mechanism of power load is analyzed, and the similar day and gray correlation analysis methods are introduced to build the load and feature fusion data set; Secondly, the nonlinear and highly fluctuating original load data is decomposed into several relatively stable components by using the variational modal decomposition (VMD) method, and the BiLSTM prediction model is built for each component; Finally, the whale optimization algorithm (WOA) is used to optimize the decomposition parameters and similar days of the model to reduce the inherent error of the model. Based on the actual data of a region in New England, the simulation results show that the MAPE, MAE and RMSE of the proposed model are 0.58%, 42 and 78 respectively, which are better than the control model and effectively improve the accuracy of load forecasting.