基于INGO-TCN-Attention的多特征短期负荷预测方法  

INGO-TCN-Attention-based Multi-feature Short-term Load Forecasting

作　　者：张晓虎 欧科宏 游鑫 黄嘉懿 ZHANG Xiaohu;OU Kehong;YOU Xin;HUANG Jiayi(College of Electrical and information Engineering,Hunan University of Technology,Zhuzhou 412000,China)

机构地区：[1]湖南工业大学电气与信息工程学院,湖南株洲412000

出　　处：《电工技术》2024年第19期17-22,共6页Electric Engineering

基　　金：国家重点研发计划项目(编号2022YFE0105200)。

摘　　要：针对人工神经网络参数随机初始化给短期电力负荷预测带来的不足,提出一种基于改进北方苍鹰算法(Improved Northern Goshawk Optimization,INGO)优化时间卷积神经网络(Temporal Convolutional Networks,TCN)融合注意力机制(Attention)的短期负荷预测方法。首先采用多策略改进北方苍鹰算法,通过基准函数测试改进前后算法的性能,表明INGO算法具有更好的寻优能力。最后,引入INGO算法对TCN进行优化,建立INGO-TCN-Attention短期电力负荷预测模型。通过实例分析和实验对比,表明INGO-TCN-Attention模型的稳定性和预测精度均优于其他模型。Aiming at addressing the shortcomings of short-term power load forecasting caused by random initialization of artificial neural network parameters,this work studied a short-term load forecasting method utilizing the temporal convolutional network optimized by improved northern goshawk optimization(INGO)and hybridized with the attention mechanism.The multi-strategy algorithm was used to improve the northern goshawk algorithm,and a test by benchmark function verified the better optimization performance of the improved algorithm.Then the INGO algorithm was further introduced to TCN,and the INGO-TCN-Attention short-term power load forecasting model was established,which in the subsequent comparative experiment exhibited stability and prediction accuracy superior to other models.

关 键 词：改进北方苍鹰算法 时间卷积神经网络 注意力机制 短期负荷预测 

分 类 号：TM411[电气工程—电器]