摘要
为满足风电场远程集控中心高效、低成本预测不同地理位置风电场风速的要求,结合“离线训练,在线预测”的思想,提出一种基于多方面特征提取和迁移学习的多变量风速预测模型。离线模型融合双通道卷积神经网络和双向长短时记忆神经网络捕捉风速信息,学习各典型位置风电场的风速特性,然后迁移至任意风电场实现快速在线预测,通过改进的多目标蝗虫优化算法集成各典型风电场预测结果,进一步提高预测精度。最后通过河北一集控中心验证表明,该文所提模型的适应性与准确性均优于其他基线模型。
In order to meet the requirements of remote control centers for efficient and low-cost wind speed prediction of wind farms at different locations,this paper proposes a multivariable wind speed prediction model based on multiple feature extraction and transfer learning by combining the idea of "offline training,online prediction".The offline model fuses wind speed information captured by twochannel convolutional neural network and bi-directional long-short-term memory neural network.Wind speed characteristics of wind farms at typical locations are learned,and then the wind speed characteristics are transferred to other wind farms to achieve online prediction.The prediction accuracy is further improved by using an improved multi-objective grasshopper optimization algorithm,which integrates the prediction results of each typical wind farm.Finally,the superiority of the model is verified by the data of a centralized control center in Hebei.The results show that the adaptability and accuracy of the proposed model are superior than that of other baseline models.
作者
梁涛
陈春宇
谭建鑫
井延伟
Liang Tao;Chen Chunyu;Tan Jianxin;Jing Yanwei(College of Artificial Intelligence and Data Science,Hebei University of Technology,Tianjin 300130,China;Hebei Construction Energy Investment Co.,Ltd.,Shijiazhuang 050051,China)
出处
《太阳能学报》
EI
CAS
CSCD
北大核心
2023年第4期132-139,共8页
Acta Energiae Solaris Sinica
基金
河北省科技支撑计划(19210108D,19214501D,20314501D,F2021202022)。
关键词
风能
风速预测
特征提取
卷积神经网络
双向长短时记忆神经网络
迁移学习
多目标蝗虫优化算法
wind energy
wind speed prediction
feature extraction
convolutional neural network
bi-directional long short-term memory network
transfer learning
multi-objective grasshopper optimization algorithm
