基于ESMD-PE和ADBN的短期电力负荷预测

王国娟; 冷建伟

王国娟,冷建伟.基于ESMD-PE和ADBN的短期电力负荷预测[J].电测与仪表,2023,60(1):29-35.
wangguojuan,lengjianwei.Short-term%20power%20load%20forecasting%20based%20on%20ESMD-PE%20and%20ADBN[J].Electrical Measurement & Instrumentation,2023,60(1):29-35.

基于ESMD-PE和ADBN的短期电力负荷预测

Short-term%20power%20load%20forecasting%20based%20on%20ESMD-PE%20and%20ADBN

DOI：10.19753/j.issn1001-1390.2023.01.005

中文关键词: 短期负荷预测 ESMD 排列熵深度信念网络自适应学习率

英文关键词:Short-term%20load%20forecasting %20ESMD %20permutation%20entropy %20deep%20belief%20network %20adaptive%20learning%20rate

基金项目:

作者中文名	作者英文名	单位
王国娟	wangguojuan	天津理工大学
冷建伟	lengjianwei	天津理工大学

摘要点击次数: 889

中文摘要:

为了改善短期电力负荷预测性能，提出了一种基于极点对称经验模式分解（ESMD）-排列熵（PE）和自适应深度信念网络（ADBN）的组合预测新方法。为了提高预测精度以及降低原始负荷序列复杂度简化预测模型输入，首先运用ESMD方法将原始负荷序列分解成为一系列复杂度互异的模态函数，然后运用排列熵计算各模态函数的熵值并对复杂度相近的模态进行重构得到新的子序列；在综合考虑各影响因素的基础上，对新序列分别构造不同的DBN预测模型，最后叠加预测结果；由于DBN模型中无监督训练阶段学习率通常采用全局统一的常数型参数，将自适应学习率引入到对比差度（CD）算法中，通过自动调整学习率改善模型的收敛速度，同时预测精度也有提高。通过算例分析，文章提出的ESMD-PE-ADBN模型的MAPE值与RMSE值分别为1.03%和90.91MW，预测效果最佳。

英文摘要:

In%20order to improve%20the%20short-term%20power%20load%20forecasting%20performance,%20a%20new%20combined%20forecasting%20method%20based%20on%20pole%20symmetrical%20empirical%20mode%20decomposition%20(ESMD)%20-permutation%20entropy%20(PE)%20and%20adaptive%20deep%20belief%20network%20(ADBN)%20is%20proposed.%20In%20order to improve%20the%20prediction%20accuracy%20and%20reduce%20the%20complexity%20of%20the%20original%20load%20sequence%20and%20simplify%20the%20input%20of%20the%20prediction%20model,%20the%20ESMD%20method%20is%20first%20used to decompose%20the%20original%20load%20sequence%20into%20a%20series%20of%20modal%20functions%20of%20different%20complexity,%20and%20then%20the%20permutation%20entropy%20is%20used to calculate%20the%20entropy%20value%20of%20each%20modal%20function%20and%20Reconstructing%20modalities%20of%20similar%20complexity to obtain%20new%20subsequences;%20on%20the%20basis%20of%20comprehensive%20consideration%20of%20various%20influencing%20factors,%20construct%20different%20DBN%20prediction%20models%20for%20the%20new%20sequence,%20and%20finally%20superimpose%20the%20prediction%20results;%20due to the%20unsupervised%20training%20stage%20in%20the%20DBN%20model%20The%20learning%20rate%20usually%20uses%20globally%20uniform%20constant%20parameters.%20The%20adaptive%20learning%20rate%20is%20introduced%20into%20the%20contrast%20difference%20(CD)%20algorithm.%20The%20convergence%20of%20the%20model%20is%20improved%20by%20automatically%20adjusting%20the%20learning%20rate,%20and%20the%20prediction%20accuracy%20is%20also%20improved.%20Through%20example%20analysis,%20the%20MAPE%20and%20RMSE%20values%20of%20the%20ESMD-PE-ADBN%20model%20proposed%20in%20the%20paper%20are%201.03%%20and%2090.91MW,%20respectively,%20and%20the%20prediction%20effect%20is%20the%20best.

查看全文查看/发表评论下载PDF阅读器

关闭