融合K近邻信息的局部三层交替优化MK诊断模型

Multi-kernel fault diagnosis model with K nearest neighbor information based on localized three-step alternating optimization

针对小样本条件下航空电子设备模块级功能故障诊断问题,将局部多核学习的局部特征表示能力与超限学习机运算高效的特点相结合,提出一种新的多故障诊断模型。一方面,通过将训练样本的近邻信息融入学习过程中,有效提升了诊断模型的推广能力;另一方面,通过构造关于局部多核超限学习机初始-对偶混合优化问题的三步交替优化策略,实现了小样本条件下对故障信息的深度挖掘,并在l1-范数约束与l2-范数约束下分别实现了对局部核权重的迭代更新。将所提诊断模型应用于某型前端接收机,结果表明,相比一般的多核、局部多核诊断方法,所提方法在实现低漏警、低虚警的同时,l1-范数约束的诊断方法将诊断精度平均提升了3.57%,l2-范数约束的诊断算法将诊断精度平均提升了4.61%。

To deal with problem of module-level functional fault diagnosis for avionics in small sample size,a new multi-fault diagnosis model is proposed in this study. The local feature representation ability of localized multiple kernel learning and the fast computing power of extreme learning machine are combined. On the one hand,the nearest neighbor information of training samples is integrated into the learning process. In this way,the scalability of the learned model is effectively improved. On the other hand,a three-step alternating optimization strategy is conducted to alternately optimize model parameters. The primal-dual optimization problem of localized multikernel extreme learning machine is utilized. The useful fault information can be deeply mined and the localized kernel weights can be independently obtained by solving their exclusive sample-wise objective function by either a l1-norm constraint or a l2-norm constraint.The proposed method is applied into the front-end receiver fault diagnosis. Compared with the state-of-art multi-kernel and localized multi-kernel diagnosis methods,experimental results show that the proposed method has lower false alarm rate and missing alarm rate.The average diagnostic accuracy is increased by 3. 57% and 4. 61% when l1-norm constraint and l2-norm constraint are used,respectively.