基于聚合时空图卷积网络的多风场超短期风速预测
Ultra-short-term Wind Speed Prediction for Multiple Wind Farms Based on Aggregated Spatio-temporal Graph Convolutional Networks
徐辰晓 1崔承刚 1郭为民 2杨宁 1刘备 2孟青叶2
作者信息
- 1. 上海电力大学自动化工程学院,上海 200090
- 2. 润电能源科学技术有限公司,郑州 450052
- 折叠
摘要
在一定环境内区域风电场呈不规则分布的条件下,传统卷积神经网络预测方法无法体现出各区域风场的分布状态和影响关系,难以实现对风速的准确预测.针对此问题,采用图卷积网络进行特征建模,并根据多风场的拓扑结构和各区域风场风速的互相关系数建立连通图和权重矩阵.其次,依赖风场风速的时间动态特征,采用改进并列式卷积结构获取同一风场下多时间段的风速序列相关性.再次,利用风场风速的空间相关性和延时效应,采用二阶聚合方法将不同区域内风速的时空特征聚合.最后,经某区域风场数据验证表明,在0～4 h预测尺度下该方法在多风场超短期风速预测中具有提取时空特征并提升预测性能的效果.
Abstract
In a certain environment where regional wind farms distribute irregularly,the traditional convolutional neural network prediction method cannot reflect the distribution states or influence relationship of regional wind farms,and it is difficult to accurately predict the wind speed.First,to solve this problem,the technology of graph convolutional networks is used for feature modeling,and the connected graph and weight matrix are established according to the topology of multiple wind farms and the cross-correlation coefficient of wind speed in each region.Second,depending on the time dynamic characteristics of wind speed at wind farms,an improved parallel convolution structure is used to obtain the correlation between wind speed series in multiple time periods at the same wind farm.Third,based on the spatial correlation and delay effect of wind speed at wind farms,the spatio-temporal characteristics of wind speed in different regions are aggregated by using a second-order aggregation method.Finally,the verification of data from one regional wind farm shows that the proposed method can extract the spatio-temporal characteristics and improve the performance of ultra short-term wind speed prediction for multiple wind farms on 0-4 h prediction scale.
关键词
风速预测/聚合时空图卷积网络/时空相关性Key words
Wind speed prediction/aggregated spatio-temporal graph convolutional networks/spatio-temporal correlation1引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据