摘要
为解决风电机组故障诊断中故障样本不足而导致模型准确率不高的问题,将当下备受关注的数据增强方法-去噪扩散概率模型(denoising diffusion probability model,DDPM)引入到故障诊断领域以生成大量高质量的故障样本数据集。因此,结合Transformer网络,提出了一种DDPM-Transformer风电机组故障样本生成方法。首先,将用于计算机视觉图像生成领域的DDPM模型应用于风电机组故障诊断领域中,通过前向加噪过程将数据逐渐转化为噪声,再通过逆向去噪过程将噪声逐步恢复为原始数据,实现从噪声中生成故障数据,解决数据不平衡问题;其次,通过对原始DDPM中使用的U-net模块进行改进,使用Transformer模型替换U-net网络,利用扩散后的数据和添加的噪声训练Transformer模型,实现噪声预测,以提高故障数据的生成质量;最后,使用多种生成模型评价指标对生成的故障数据进行评价,在监督控制和数据采集系统(supervisory control and data acquisition,SCADA)故障数据生成中论证改进DDPM-Transformer模型的性能。通过试验证明,所提DDPM-Transformer模型与现有的生成模型相比,最大均值异(maximum mean discrepancy,MMD)最大提升0.13,峰值信噪比(peak signal to noise ratio,PSNR)最大提升7.8。所提模型可以有效地生成质量更高的风电机组故障样本,从而基于该样本集辅助训练基于深度学习的故障诊断模型,可以使诊断模型具有更高精度和良好的稳定性。
In order to solve the problem that the model accuracy is not high due to insufficient fault samples in the fault diagnosis of wind turbine,the current data enhancement method,denoising diffusion probability model(DDPM),was introduced into the fault diagnosis field to generate a large number of high⁃quality fault sample data sets.Therefore,with Transformer network,a DDPM⁃Transformer wind turbine fault samples generation method was proposed.Firstly,the DDPM model used in the field of computer vision image generation was applied to the field of wind turbine fault diagnosis,where the data was gradually transformed into noise through the forward noise addition process,and then the noise was progressively restored to the original data through the inverse denoising process,so as to realize the generation of fault data from noise and solve the problem of data imbalance.Secondly,by improving the U⁃net module used in the original DDPM and replacing the U⁃net network with the Transformer model,the noise prediction was realized by training the Transformer model using the diffused data and the noise to improve the quality of fault data generation.Finally,a variety of generation model evaluation indexes were used to evaluate the generated fault data,and the improved performance of DDPM⁃Transformer model was demonstrated in supervisory control and data acquisition(SCADA)fault data generation.Experimental results show that compared with existing generation models,the proposed DDPM⁃Transformer model has a maximum increase of 0.13 maximum mean discrepancy(MMD)and 7.8 peak signal to noise ratio(PSNR).The model in this paper can effectively generate higher⁃quality wind turbine fault samples,so that the fault diagnosis model based on deep learning can be trained based on the sample set,which can make the diagnosis model have higher accuracy and good stability.
作者
孟昱煜
张沣琦
火久元
常琛
陈峰
MENG Yuyu;ZHANG Fengqi;HUO Jiuyuan;CHANG Chen;CHEN Feng(School of Electronic and Information Engineering,Lanzhou Jiaotong University,Lanzhou 730070,China;National Cryosphere Desert Data Center,Northwest Institute of Eco-Environment and Resources,Chinese Academy of Sciences,Lanzhou 730000,China)
出处
《振动与冲击》
北大核心
2025年第4期286-297,共12页
Journal of Vibration and Shock
基金
国家自然科学基金(62262038)
甘肃省重点研发计划-工业项目(22YF7GA145)。
关键词
DDPM
TRANSFORMER
风电机组
故障诊断
样本生成
denoising diffusion probability model(DDPM)
Transformer
wind turbines
fault diagnosis
sample generation
