方娜, 李俊晓, 陈浩, 李新新. 基于变分模态分解的卷积神经网络−双向门控循环单元−多元线性回归多频组合短期电力负荷预测[J]. 现代电力, 2022, 39(4): 441-448. DOI: 10.19725/j.cnki.1007-2322.2021.0130
引用本文: 方娜, 李俊晓, 陈浩, 李新新. 基于变分模态分解的卷积神经网络−双向门控循环单元−多元线性回归多频组合短期电力负荷预测[J]. 现代电力, 2022, 39(4): 441-448. DOI: 10.19725/j.cnki.1007-2322.2021.0130
FANG Na, LI Junxiao, CHEN Hao, LI Xinxin. Multi-Frequency Combination Short-term Power Load Forecasting with Convolutional Neural Networks - Bidirectional Gated Recurrent Unit-Multiple Linear Regression based on Variational Mode Decomposition[J]. Modern Electric Power, 2022, 39(4): 441-448. DOI: 10.19725/j.cnki.1007-2322.2021.0130
Citation: FANG Na, LI Junxiao, CHEN Hao, LI Xinxin. Multi-Frequency Combination Short-term Power Load Forecasting with Convolutional Neural Networks - Bidirectional Gated Recurrent Unit-Multiple Linear Regression based on Variational Mode Decomposition[J]. Modern Electric Power, 2022, 39(4): 441-448. DOI: 10.19725/j.cnki.1007-2322.2021.0130

基于变分模态分解的卷积神经网络−双向门控循环单元−多元线性回归多频组合短期电力负荷预测

Multi-Frequency Combination Short-term Power Load Forecasting with Convolutional Neural Networks - Bidirectional Gated Recurrent Unit-Multiple Linear Regression based on Variational Mode Decomposition

  • 摘要: 为了有效提高电力负荷预测精度,针对电力负荷非线性、非平稳性、时序性的特点,提出了一种卷积神经网络(convolutional neural networks,CNN)、双向门控循环单元(bidirectional gated recurrent unit,BiGRU)和多元线性回归(multiple linear regression,MLR)混合的多频组合短期电力负荷预测模型。该模型先利用关联度分析得到相似日,并将其负荷组成新的数据序列,同时使用变分模态分解(variational mode decomposition,VMD)将该数据序列进行分解,并重构成高低2种频率。对于高频分量,使用CNN-BiGRU模型进行预测;低频部分则使用MLR。最后将各个模型得出的预测结果叠加,得到最终预测结果。以2006年澳大利亚真实数据为例,进行短期电力负荷预测。仿真结果表明,相比于其他网络模型,该模型具有较高的预测精度和拟合能力,是一种有效的短期负荷预测方法。

     

    Abstract: To effectively improve the accuracy of power load forecasting and in allusion to such characteristics of power load as nonlinearity, non-stationary and time sequence, a multi-frequency combination power load forecasting model, in which the Convolutional Neural Network (abbr. CNN) and the Bidirectional Gated Recurrent Unit (abbr. BiGRU) and the Multiple Linear Regression (abbr. MLR were mixed, was proposed. Firstly, in the proposed model the correlation degree analysis was utilized to obtain similar days and their loads were constituted new data series, meanwhile the variational mode decomposition (abbr. VMD) was used to decompose the obtained data series and reconstruct into high and low frequencies. As for the high-frequency component the CNN-BiGRU model was used for the prediction; and for the low-frequency component the MLR was used. Finally, superposing the predicted results obtained by above mentioned two models the final predicted results could be obtained. Based on the real data of Australia in 2006, a short-term load forecasting was performed. Simulation results show that comparing with other network models, by use of the proposed model the forecasting results possess higher prediction accuracy and fitting ability.

     

/

返回文章
返回