基于CEEMDAN-SE-VMD和CNN-BIGRU的短期负荷预测

Short-Term Load Forecasting Based on CEEMDAN-SE-VMD and CNN-BIGRU

针对电力负荷预测的精度较低问题,提出一种基于CEEMDAN-SE-VMD和CNN-BIGRU组合模型的负荷预测方法。首先该模型采用自适应噪声的完全经验模态分解(CEEMDAN)处理成分复杂的原始负荷数据,经过分解后得到若干个包含不同频率成分的本征模函数(IMF)。再利用样本熵(SE)对分解后不同频率的本征模函数进行熵值聚类重组。然后,利用变分模态分解(VMD)对重组后的高频序列进行二次分解,将二次分解后得到的子序列和样本熵重组的低频序列和趋势序列数据输入卷积神经网络(CNN)网络,利用其来提取反映负荷序列复杂相关的高位特征向量。最后,再输入到双向门控循环单元(BIGRU)网络中进行预测,得到各子序列的预测结果,叠加得到最终的负荷序列预测结果。通过横向和纵向实验结果对比,证明所提出的模型能够较好地提升电力负荷预测精度。

Targeting at the problem of low accuracy in electricity load forecasting,a load forecasting method based on CEEMDAN-SE-VMD and CNN-BIGRU combined model is proposed.Firstly,the model first applies complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)to decompose the complex raw load data.,and several intrinsic mode functions(IMF)with different frequency components are obtained after decomposition.Then sample entropy(SE)is used to cluster the decomposed intrinsic mode functions with different frequencies.Then,variational mode decomposition(VMD)is used to decompose the reconstructed high frequency sequences.The sub-sequences obtained from the quadratic decomposition and the low frequency sequence and trend sequence data reconstructed by sample entropy(SE)are input into CNN network to extract the high-level feature vectors reflecting the complexity and correlation of load sequences.Finally,it is input into the bidirectional gated recurrent unit(BIGRU)network for prediction,the prediction results of each subsequence are obtained,and the final load sequence prediction results are superimposed.Through the comparison of transverse and longitudinal experimental results,it is proved that the proposed model can improve the accuracy of power load prediction.