基于图注意力网络的短期负荷预测模型  

Short-term load forecasting model based on graph attention network

作　　者：何群英[1] 谷卫 HE Qun-ying;GU Wei(School of Mechanical and Electrical Engineering,Zhejiang Tongji Vocational College of Science and Technology,Hangzhou 311231,China;School of Automation,Nanjing University of Information Science and Technology,Nanjing 210044,China)

机构地区：[1]浙江同济科技职业学院机电工程学院,浙江杭州311231 [2]南京信息工程大学自动化学院,江苏南京210044

出　　处：《水电站机电技术》2024年第7期40-46,146,共8页Mechanical & Electrical Technique of Hydropower Station

基　　金：浙江同济科技职业学院科研项目(FRF23YB009);国家自然科学基金资助项目(62172363)。

摘　　要：现有方法在处理负荷数据的非线性、动态性以及空间和时间依赖性方面存在局限,导致预测精度较低。为解决上述问题,本文提出了一种基于改进变分模态分解(SVMD)、图注意力网络(GAT)和长短期记忆网络(LSTM)的混合模型,并通过白鲸优化算法(BWO)进行参数优化。该模型旨在通过综合各组件的优势,提高短期负荷预测的准确性和鲁棒性。模型首先应用SVMD对负荷数据进行有效分解,提取关键模态分量;接着,GAT网络捕捉负荷波动性和加权空间的相关性,增强模型对负荷数据空间特征的感知;LSTM网络提取和预测时间序列特征;最后,用BWO算法优化模型参数,提高模型的收敛速度和预测性能。试验结果表明,SVMD-GAT-BWO-LSTM模型在平均绝对百分比误差(E_(MAPE))、均方根误差(E_(RMSE))和拟合度(R^(2))等评价指标上均表现最佳,显示出较高的稳定性和鲁棒性。消融试验进一步证实了模型中每个组件的重要性和对提高预测性能的贡献。Existing methods have limitations in dealing with the nonlinearity,dynamic nature,and spatial and temporal dependence of load data,resulting in low forecast accuracy.In order to solve the above problems,this paper proposes a hybrid model based on the sequential variational mode decomposition(SVMD),graph attention network(GAT)and long short-term memory network(LSTM),and optimizes the parameters through beluga whale optimization(BWO).The model aims to improve the accuracy and robustness of short-term load forecasting by combining the advantages of all components.For the model,SVMD was first applied to effectively decompose load data and extract key modal components;then,GAT captured the correlation between the load fluctuation and weighted space to enhance the perception of the model on spatial characteristics of load data;LSTM extracted and forecasted time series characteristics;finally,BWO was used to optimize model parameters and improve the convergence speed and forecast performance of the model.The test results show that the SVMD-GAT-BWO-LSTM model performs best in terms of the mean absolute percentage error(E_(MAPE)),root mean square error(E_(RMSE))and degree of fitting(R^(2))and shows high stability and robustness.The ablation tests further verified the importance of all components in the model and their contributions to improving forecast performance.

关 键 词：负荷预测 图注意力 SVMD 白鲸优化 LSTM 

分 类 号：TM715[电气工程—电力系统及自动化]