基于多协同映射核Fisher判别分析的多维状态监测数据特征提取方法OA

Conventional kernel-based Fisher discriminant analysis(KFDA)is limited in adaptability of kernel functions for feature extraction from condition monitoring data.To address this,this paper proposes a feature extraction method based on multiple co-mapping kernel Fisher discriminant analysis(MCKFDA),which incorporates kernel function optimization into the conventional framework.Firstly,a data-dependent kernel function is constructed to adapt the kernel structure to various types of condition monitoring data collected;on this basis,a multiple co-mapping kernel structure is devised to accommodate diversity across data samples.Then,the maximun margin criterion(MMC)is employed to achieve the optimal combination weights of the multiple co-mapping kernel,while the Fisher criterion is used to find the optimal data-dependent kernel parameters.The proposed approach was assessed using a lithium-ion battery dataset.Experimental results show that,compared with traditional kernel discriminant analysis,the proposed MCKFDA method improves the mean square error by 15.4%for lithium-ion battery capacity estimation,representing a substantial enhancement in accuracy of performance evaluation.