针对不同负荷工况下,热工参数数据分布差异大且故障类别不一致的问题,提出了一种基于多源样本加权域对抗网络的热力系统故障诊断方法.首先,构建领域共享的一维卷积神经网络以提取多个源域和目标域的深度判别特征;其次,引入加权机制和域一致性损失度量样本,以降低仅存在于源域的故障类别的负迁移影响;然后,通过多域判别器的对抗学习实现每对源域和目标域的特征差异对齐;最后,构建多分类器对齐模块以提高预测的一致性,从而实现多源域不同工况下热力系统故障的准确诊断.借助某 600 MW 超临界机组全范围仿真系统进行故障仿真实验,结果验证了所提方法的鲁棒性和优越性.
In order to solve the problem of large distribution differences in thermal parameters and inconsistent fault categories under different loading conditions,a thermal system fault diagnosis method based on multi-source sample weighted domain adversarial network is proposed.Firstly,a domain-shared one-dimensional convolutional neural network is constructed to extract the deep discriminate features of multiple source domains and target domain.Secondly,the weighted mechanism and the domain consistency loss measure samples are introduced to reduce the negative transfer impact of fault categories that exist only in the source domains.Furthermore,the feature difference alignment between each pair of source and target domain is achieved by adversarial learning of multi-domain discriminators.Finally,a multi-classifier alignment module is constructed to improve the consistency of predictions,so as to realize the accurate fault diagnosis of thermal system under different working conditions in multiple-source domains.The fault simulation experiments are carried out on a fullscope simulator of a 600 MW power unit,and the results verify the robustness and superiority of the proposed method.