An Effective Kernel Online Learning Algorithm with Sparse Update for Electronic Devices Condition Prediction

Abstract

Abstract In order to implement condition prediction for electronic devices, a new online learning algorithm for extreme learning machine with kernel (KELM) is proposed. For key nodes selection, a sparse dictionary with predefined size is constructed adaptively by online minimization of its cumulative coherence. Meanwhile, for model coefficients update, an improved decremental learning algorithm is presented by using elementary transformation of Gram matrix and block matrix inversion formula. The performance of proposed algorithm is compared with several well-known online sequential ELM algorithms. The simulation results show that the proposed algorithm has higher prediction accuracy and better stability. Thus it is suited to online condition monitoring of electronic devices.