基于损失优化的DAE降低风电机组不确定因素的影响

Reducing the Impact of Uncertain Factors on Wind Turbine Units Based on Loss-Optimized DAE

摘要: 考虑到风电机组运行过程中存在大量的不确定因素，这些不确定因素会导致监测参数出现质量波动，进而影响风电机组的异常检测结果或状态评估。为了解决不确定因素对风电机组监测参数的影响，提出一种基于损失优化的深度自编码器模型，降低风电机组中不确定因素的影响。该模型结合信息瓶颈理论，并以最小充分统计变量的存在条件为损失函数来训练。从而实现在对数据压缩重构的过程中，去掉不确定因素带来的干扰信息，同时最大程度上地保留有用的信息，以降低监测参数的波动。通过实际案例表明该方法的有效性，为后续的高精度的异常状态检测和状态评估奠定良好的基础。

Abstract: Considering that there are a large number of uncertain factors in the operation of wind turbines, these uncertain factors will lead to quality fluctuations in monitoring parameters, which will affect the abnormal detection results or status evaluation of wind turbines. In order to solve the influence of uncertain factors on the monitoring parameters of wind turbines, a deep autoencoder model based on loss optimization is proposed to reduce the influence of uncertain factors in wind turbines. The model incorporates information bottleneck theory and is trained with the existence condition of the minimum sufficient statistical variable as the loss function. In this way, in the process of data compression and reconstruction, the interference information caused by uncertain factors is re-moved, while the useful information is retained to the greatest extent, so as to reduce the fluctua-tion of monitoring parameters. The effectiveness of the method is shown through practical cases, which lays a good foundation for subsequent high-precision abnormal state detection and state evaluation.