基于重构分解和误差补偿的短期电力负荷双层协同预测

Bi-Level Collaborative Short-Term Power Load Forecasting Based on Reconstruction Decomposition and Error Compensation

中文关键词:短期电力负荷  重构分解  误差补偿  双层协同预测

keywords:short-term power load  reconstruction decompo-sition  error compensation  bi-level collaborative forecasting

中文摘要:为有效提高短期电力负荷预测精度，提出了一种基于重构分解和误差补偿的双层协同预测方法。上层以改进的完全集合经验模态分解（improved complete ensemble empirical mode decomposition with adaptive noise，ICEEMDAN)）、排列熵、K-medoids聚类和自适应变分模态分解组成的重构分解方法降低负荷序列的不可预测性，进一步构建时间卷积网络联立双向门控循环单元的混合预测模型（temporal convolutional network-bidirectional gated recurrent unit，TCBG）；下层以上层负荷预测误差以及历史误差为输入，建立ICEEMDAN-TCBG误差补偿模型，修正上层预测结果。以爱尔兰地区2019年电力负荷为算例搭建多组分析实验，实验结果表明：所提方法的MAE和MAPE分别为298.079 MW和1.242%，优于其他对比方法。

Abstract:To effectively improve the accuracy of short-term power load forecasting， this paper proposes a bi-level collaborative forecasting method based on reconstruction decomposition and error compensation. In the upper layer， the reconstruction decomposition method composed of improved complete ensemble empirical mode decomposition with adaptive noise（ICEEMDAN），permutation entropy， K-medoids clustering and adaptive variational mode decomposition reduces the unpredictability of load sequence， and further constructs a hybrid prediction model of temporal convolutional network combined with bidirectional gated recurrent unit （TCBG）. The lower layer takes the upper layer load forecast error as well as the historical error as inputs， and builds the ICEEMDAN-TCBG error compensation model to correct the upper tier forecast results. Taking the power load of Ireland in 2019 as an example， the results show that the MAE and MAPE of the proposed method are 298.079 MW and 1.242%， respectively， which are better than other comparison methods.