基于隶属度和LMK-ELM的航空电子部件诊断方法

为提高航空电子部件模块级故障诊断精度，提出一种新的面向"软聚类"的局部多核学习（LMKL）-超限学习机（ELM）离线诊断方法。通过引入模糊C均值聚类对样本空间进行模糊划分，挖掘聚类内部多样性的同时，实现了对过学习的抑制；将模糊划分产生的隶属度信息融入LMKL-ELM的优化过程，运用基于初始-对偶混合优化问题的三步优化策略克服了局部核权重二次非凸的问题，在l₁-范数与l₂-范数约束下分别给出了相应的更新方法。将所提方法应用于某型机前端接收机，结果表明：与4种流行的多核诊断方法相比，该方法可有效避免漏警、抑制虚警，在l₁-范数和l₂-范数约束下，其诊断精度比其他方法的平均值分别提升了4.09%和5.13%。

Diagnosis method for avionics based on membership and LMK-ELM

To improve the accuracy of module-level fault diagnosis for avionics, a new off-line diagnosis method based on soft-clustering-sensitive Localized Multi-Kernel Learning (LMKL) and Extreme Learning Machine (ELM) is proposed. By introducing fuzzy C-means clustering to partition the sample space, the over-learning is suppressed while mining the diversity within the cluster. The membership information generated by the fuzzy partition is integrated into the optimization process of LMKL-ELM. A three-step optimization strategy based on the initial-dual hybrid optimization problem is used to overcome the quadratic non-convexity of the local kernel weights. The corresponding updating methods for these weights are given under l₁-norm constraint and l₂-norm constraint. The proposed method is applied to the front-end receiver. Compared with four popular multi-kernel diagnostic algorithms, the results show that the proposed method can effectively avoid missing alarm and suppress false alarm. The diagnostic accuracy is 4.09% higher in l₁-norm and 5.13% higher in l₂-norm than the average of other methods.