基于Attention机制优化GRU混合神经网络的配电网负荷预测模型应用  

Distribution Network Load Forecasting Based on Attention-mechanism-optimized GRU Hybrid Neural Network

作　　者：蒋军 JIANG Jun(State Grid Hunan Electric Power Co.,Ltd.Shaoyang Power Supply Branch,Shaoyang 422000,China)

机构地区：[1]国网湖南省电力有限公司邵阳供电分公司,湖南邵阳422000

出　　处：《电工技术》2024年第19期23-25,28,共4页Electric Engineering

摘　　要：随着全球能源需求的持续增长和环境保护意识的增强,配电网负荷预测成为电力系统运行和规划中的关键环节。为此提出了一种基于Attention机制优化的GRU混合神经网络模型,进一步提升配电网负荷预测的精度和鲁棒性。通过对传统GRU模型和引入Attention机制的GRU模型进行比较,表明带Attention机制的GRU模型在预测精度和稳定性方面表现更优。在公开数据集上的测试结果显示,带Attention机制的GRU模型显著降低了预测误差,提高了预测精度,为智能配电网的高效运行提供了有力支持。The continuous growth of global energy demand and the increasing awareness of environmental protection have made load forecasting in distribution networks a crucial part in operation and planning of power systems.The present work made a preliminary attempt to establish an attention-mechanism-optimized GRU hybrid neural network model to improve accuracy and robustness of distribution network load forecasting.The proposed model was verified by a test on public database,compared with conventional GRU model,superior in forecasting accuracy and stability,and thereby potentially conducive to efficient operation of smart distribution networks.

关 键 词：配电网 负荷预测 GRU神经网络 Attention机制 深度学习 时间序列预测 

分 类 号：TM715[电气工程—电力系统及自动化]