Aeroengines Remaining Useful Life Prediction Based on Improved C-Loss ELM

Abstract

A correntropy induced loss (C-loss) function is a loss function developed based on entropy theory. Due to the non-convexity of the C-loss function, ELM based on C-loss has been proven to have better regression effects and robustness. Inspired by generalized quantile learning, we find that there is a further improvement in the C-loss function. The main work of this paper is to propose a new improved C-loss function, which lead to a novel algorithm ICELM, and the half-optimized algorithm is used to solve the ICELM. Experiments on several benchmark data sets prove the superiority of the ICELM. Compared with CELM, ICELM has more efficient regression performance and robustness, and is obviously more suitable for complex aeroengines remaining useful life prediction field which affected by noise and outliers. Experiments on a benchmark C-MAPSS dataset also validate the theory.