基于图卷积神经网络−双向门控循环单元及注意力机制的风电功率短期预测模型  

A Short-term Wind Power Prediction Model Based on Graph Convolutional Neural Network-bidirectional Gated Recurrent Unit and Attention Mechanism

作　　者：张光昊 张新燕[1] 王朋凯 ZHANG Guanghao;ZHANG Xinyan;WANG Pengkai(College of Electrical Engineering,Xinjiang University,Urumqi 830047,Xinjiang Uygur Autonomous Region,China)

机构地区：[1]新疆大学电气工程学院,新疆维吾尔自治区乌鲁木齐市830047

出　　处：《现代电力》2025年第2期201-208,共8页Modern Electric Power

基　　金：国家自然科学基金项目(51667018);新疆维吾尔自治区自然科学基金项目(2021D01C044)。

摘　　要：风电功率的准确预测对电力系统的稳定运行意义重大,针对传统组合模型难以充分挖掘变量间潜在依赖性,导致在高维度、大量数据下风电功率预测精度偏低的问题。该文提出一种图卷积神经网络–双向门控循环单元及注意力机制的短期风电功率预测模型。该模型以数值天气预报数据(numerical weather prediction,NWP)和风电功率历史数据作为输入,首先利用皮尔逊相关性分析筛选特征,然后借助残差连接的图卷积神经网络(graph convolutional neural network,GCN)和图学习层挖掘空间特征关系,接着采用双向门控循环单元(bidirectional gated recurrent unit,BiGRU)挖掘历史数据的时序特征,最后引入注意力机制(attentional mechanisms,AM)分配权重,实现风电功率短期预测。以某风电场实测数据为例进行算例分析,实验结果表明,该文方法在单步及多步预测中相比其他方法有更好的预测精度。Accurate prediction of wind power is of great significance to the stable operation of power systems.To address the issue that traditional combinatorial models face the challenges in fully exploring the potential dependencies among variables and exhibit low accuracy in wind power prediction when dealing with high-dimension and largescale data,a model,incorporating a graph convolutional neural network-bidirectional gated cyclic unit and an attention mechanism,is proposed for shortterm wind power prediction.The model takes numerical weather prediction(NWP)data and wind power historical data as input.It initially employs Pearson correlation analysis to filter features,followed by the utilization of graph convolutional neural network(GCN)with the help of residual connection to mine spatial feature relationships via network and graph.Subsequently,a bidirectional gated recurrent unit(BiGRU)is employed to mine the time-series features of historical data.Finally,an attention mechanism(AM)is introduced to assign weights to achieve short-term wind power prediction.The experimental results demonstrate that this method exhibits superior prediction accuracy in both single-step and multi-step prediction compared to other methods.

关 键 词：风电功率预测 混合深度神经网络 图卷积神经网络 双向门控循环单元 注意力机制 

分 类 号：TM614[电气工程—电力系统及自动化]