黄元行 刘晓波 范津玮 熊焱 叶琦



摘 要:为提高非平稳性电力负荷的预测精度以及充分挖掘历史负荷数据的时序特征,本文提出了基于注意力(Attention)机制和经验模态分解(EMD)以及门控循环神经网络(GRU)组合的负荷预测方法。首先使用EMD对原始负荷数据进行EMD分解,继而得到有限个具有本征模态函数(IMF)的分量;然后考虑到各分量间的相关性,使用多层GRU网络对IMF分量进行多输入多输出预测,同时引入注意力机制,深入挖掘历史负荷数据的时序相关性特征;最后对多层神经网络输出的有限个分量预测结果进行重构得到最终的负荷预测值。算例分析采用贵州电网某地实际负荷数据,经过与不同模型进行预测误差分析和对比,本文所提方法能够有效提高预测精度。
关键词: 电力负荷预测; 经验模态分解; 门控循环神经网络; 注意力机制
文章编号: 2095-2163(2021)03-0033-06 中图分类号: TP18 文献标志码: A
【Abstract】In order to improve the prediction accuracy of non-stationary electric load and fully exploit the time-series characteristics of historical load data, this paper proposes a method of load prediction based on a combination of Attention mechanism and empirical modal decomposition (EMD) and gated recurrent unit (GRU). Firstly, EMD is used to decompose the original load data, and then some components with intrinsic modal functions (IMFs) are obtained; subsequently, considering the correlation among these components, a multi-layer GRU network is used to predict the IMF components with multiple inputs and multiple outputs, and the Attention mechanism is also introduced to deeply explore the time-series correlation characteristics of the historical load data; finally, the results of load prediction are obtained by reconstructing the finite components from the multilayer neural network. The actual load data of a place in Guizhou power grid are used to the simulative analysis, and the prediction errors of different models are compared, which shows the proposed method could effectively improve the prediction accuracy.
【Key words】 power load forecasting; empirical modal decomposition; gated recurrent unit; Attention mechanism
0 引 言
隨着智能电网的发展,配电侧用电负荷呈现多元化,电网负荷波动性和尖峰性较为明显,其对电网的调度和运行规划提出了更高的要求。因此,准确而实时的负荷预测有助于电网的经济和可靠运行。同时随着调度系统对负荷的采集频率和数据准确性也不断提升,为负荷特性分析和负荷预测提供了高质量、海量化的数据基础[1]。
过去由于预测算法的限制以及计算机并行处理的能力和电力负荷大数据的不足,负荷预测通常采用经典的预测方法如回归分析法[2]、时间序列法[3]和传统的负荷求导法、指数平滑法[4]、卡尔曼滤波法[5]以及灰色预测法[6]等。……