【摘要】 针对极端气候条件下电网负荷预测偏差较大的问题,本文提出一种基于熵选两阶段重构与门控循环单元(CESGMD-WPTD-GRU)的负荷预测模型。首先,采用自适应噪声集合经验模态分解(CEEMDAN)对历史负荷数据进行初次分解,并引入样本熵(SampEn)对初次分解所得分量进行筛选;其次,利用辛几何模态分解(SGMD)对高熵值分量进行二次分解,并结合小波包阈值去噪技术(WPTD)以降低模态复杂度;最后,构建门控循环单元(GRU)网络捕捉以负荷时序规律。为验证模型的有效性,选取江苏省无锡市和福建省泉州市在极端天气条件下的电力负荷数据集进行实证分析。实验结果表明:相较于传统预测模型,本研究所提模型在测试集的R~2值分别达到98.47%与99.63%,为高波动负荷预测提供了有效的解决方案。更多还原

【Abstract】 Aiming at the problem of prediction bias in power grids under extreme climate,a load prediction model based on entropy-selected double-stage reconstruction and gated recurrent unit(CESGMD-WPTD-GRU) is proposed.First,the primary decomposition of historical load data is carried out by complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN),and then the sample entropy(SampEn) is introduced to filter the primary component.Then the secondary decomposition is implemented by symplectic geometry mode decomposition(SGMD) for the high entropy-valued components,followed by the wavelet packet threshold denoising(WPTD)to reduce the modal complexity.Finally,a gated recurrent unit(GRU) is employed to capture the temporal dependencies of the processed sequences.To validate the effectiveness of the proposed approach,empirical analysis are conducted using the power load dataset of Wuxi(Jiangsu Province)and Quanzhou(Fujian Province) under extreme weather scenarios.Experimental results show that compared with the traditional forecasting model,the R~2 values of this scheme in the test set reach 98.47% and 99.63%,respectively,providing an effective solution for high-volatility load prediction.更多还原