基于多级特征提取框架的风电机组载荷预测方法
LOAD PREDICTION METHOD OF WIND TURBINE BASED ON MULTISTAGE FEATURE EXTRACTION FRAMEWORK
岳健 1史秉帅 1范寒 1张克 1张海龙1
作者信息
- 1. 北京金风慧能技术有限公司,北京 100176
- 折叠
摘要
该文研究了对风电机组进行载荷预测的问题,主要从两个方面展开:SCADA数据增强与使用多级特征提取框架做载荷预测.首先采用生成对抗网络(WGAN-GP)进行数据增强.在载荷预测方面,不同于传统的Transformer模型应用于文本数据,该文使用风电机组运行时的结构化数据,且为提高特征提取能力,提出一种多级特征提取器进行特征提取.最后使用改进的Transformer模型和DNN、ResNet等模型的结果进行对比,发现多级特征提取模型对于与目标特征相关性较高的数据有较好的预测效果,同时对于相关性较低的数据也具有较好的非线性提取能力.
Abstract
In this paper,load prediction for wind turbines is studied from two aspects:SCADA data enhancement and multi-stage feature extraction framework for load prediction.Firstly,Generative adversarial network(WGAN-GP)is used for data enhancement.In terms of load prediction,different from the traditional Transformer model applied to text data,this paper uses the structured data of wind turbine operation,and in order to improve feature extraction capability,a multistage feature extractor is proposed for feature extraction.Finally,the improved Transformer model is compared with the results of DNN,ResNet and other models,and it is found that the multistage feature extraction model has better prediction effect for data with high correlation with target features,and has better nonlinear extraction ability for data with low correlation.
关键词
风电机组/Transformer/特征提取/生成对抗网络/载荷预测/数据增强Key words
wind turbines/Transformer/feature extraction/generative adversarial network/load prediction/data enhancement引用本文复制引用
出版年
2024
NETL
NSTL
万方数据