摘要
为提高风电功率预测精度及计算速度,提出了一种基于多头注意力机制卷积预测模型;利用多头注意力机制的集成作用,在继承了自注意力机制优势的同时防止了过拟合。首先,对收集到的数据进行皮尔逊相关性分析,挑选出与风电功率相关性大的变量,将筛选后的数据归一化处理并划分为训练集、验证集和测试集;然后,利用划分好的数据集对模型开展实验。实验结果表明,模型的预测误差稳定在0.05%,预测精度优于卷积神经网络、长短期记忆网络等模型。
In order to improve the prediction accuracy and calculation speed of wind power,a convolution prediction model based on multi-head attentional mechanism was proposed.With using the integration function of multi-head attentional mechanism,it inherits the advantages of self-attentional mechanism and avoids over-fitting.Firstly,Pearson correlation analysis was carried out on the collected data,and variables with high correlation with wind power were selected.The screened data were normalized and divided into training set,verification set and test set.Then,experiments are carried out on the model with using the partitioned data set.The experimental results show that the prediction error of the model is stable at 0.05%,and the prediction accuracy is better than that of convolutional neural network and long and short term memory network.
作者
李俊卿
胡晓东
秦静茹
张承志
LI Junqing;HU Xiaodong;QIN Jingru;ZHANG Chengzhi(Department of Electrical Engineering,North China Electric Power University,Baoding 071003,China)
出处
《电力科学与工程》
2022年第7期34-40,共7页
Electric Power Science and Engineering
关键词
风力发电
功率
预测分析
多头注意力机制
卷积神经网络
wind power generation
power
predictive analysis
multi-head attentional mechanism
convolutional neural network