基于多重注意力的双通道超短期风电功率预测  

Dual-channel Ultra-short-term Wind Power Prediction Based on Multi-attention

作　　者：赵雅婷 景超 张兴忠[1] ZHAO Yating;JING Chao;ZHANG Xingzhong(College of Software,Taiyuan University of Technology,Jinzhong 030600,Shanxi Province,China;College of Artificial Intelligence,Xi’an Jiaotong University,Xi’an 710049,Shaanxi Province,China)

机构地区：[1]太原理工大学软件学院,山西省晋中市030600 [2]西安交通大学人工智能学院,陕西省西安市710049

出　　处：《电网技术》2023年第7期2887-2896,共10页Power System Technology

基　　金：国网山西省电力公司科技项目(52053020000W)。

摘　　要：风电功率的超短期精确预测对于电力系统持续、稳定运行具有重要意义。针对风电功率超短期预测问题,提出了一种基于多重注意力双通道模型(dual-channel model with multi-attention,DCMMA)的预测方法,首先采用最小冗余最大相关性(minimum redundancy maximum relevance,mRMR)算法对多元气象因素变量进行筛选,预处理得到适配模型的样本;其次通过DCMMA模型并行提取气象因素时序数据和风电功率时序数据在超短期内的内在特征,并添加多重注意力对各维度下的重要信息进行关注;最后将贝叶斯优化算法(Bayesian optimization algorithm,BOA)融入模型超参数寻优过程,得到包含近似最优超参数的DCMMA模型。多次实验结果表明,所提方法在超短期内的预测精度均优于其他对比模型。The accurate prediction of the ultra-short-term wind power is of great significance for the continuous and stable operation of a power system.Aiming at this problem,a method based on the dual-channel model with the multi-attention(DCMMA)is proposed.Firstly,the minimum redundancy maximum relevance(mRMR)algorithm is used to select the multiple variables of the meteorological factors,and the adapted sample of the model is obtained by preprocessing these variables.Secondly,the DCMMA model is used to parallelly extract the internal characteristics of the meteorological factor time series data and the wind power time series data in the ultra-short term,and the multiple attentions are added to pay attention to the important information in each dimension.Finally,the Bayesian optimization algorithm(BOA)is integrated into the process of the model hyper-parameter optimization,and the DCMMA model with the approximately optimal hyper-parameters is obtained.The experimental results show that the prediction accuracy of the proposed method is higher than the other contrast models in the ultra-short term.

关 键 词：风电功率预测 双通道 多重注意力 多元气象因素 

分 类 号：TM614[电气工程—电力系统及自动化]