【摘要】 为提高风力发电中同步电机绕组短路故障特征提取能力，解决同步电机运行工况变化引起监测数据分布差异和待诊断样本标签不足的问题，提出一种基于领域自适应改进胶囊网络（DA-ICN）的同步电机故障诊断方法。首先利用改进后的胶囊网络提取源域和目标域中同步电机数据的带有空间信息的故障特征，其次结合领域自适应方法将不同工况中的特征映射到公共特征空间，随后通过改进局部最大均值距离（ILMMD）方法消除不同工况同类故障特征间的分布差异，最终得到跨工况下的同步电机故障诊断模型。经实验室数据验证，DA-ICN模型在各类跨工况实验中的平均诊断精度达到98.44%，特征提取能力、诊断精度和迁移性均优于其他方法，并且能在工况差异较大时有效实现同步电机的跨工况故障诊断。更多还原

【Abstract】 To enhance the extraction of short circuit fault characteristics in synchronous generators used in wind power, this paper introduces a Domain Adaptive Improved Capsule Network(DA-ICN) method tailored for synchronous generator fault diagnosis. This method leverages an improved capsule network to capture fault characteristics enriched with spatial information from generator data across different domains. It then utilizes domain adaptive techniques to align features from varying operating conditions into a unified feature space. The Improved Local Maximum Mean Discrepancy(ILMMD) approach is applied to mitigate the distribution discrepancies of similar fault characteristics under diverse conditions. The culmination of this process is a robust fault diagnosis model that functions effectively across different operating conditions. Experimental results using laboratory data validate the effectiveness of the DA-ICN model, which achieves an impressive average diagnostic accuracy of 98.44% in tests across different operating conditions. The proposed method not only demonstrates superior feature extraction, diagnostic precision, and adaptability compared to existing methods but also effectively handles significant disparities in operating conditions, ensuring reliable fault diagnosis in synchronous generators.更多还原