Research on a Stacking Ensemble Model With Adaptive Feature Weighting for Predicting the Tribological Properties of Lubricating Grease

Abstract

To optimize traditional stacking ensemble method and accurately predict the tribological properties of lubricating grease, this study proposed a stacking ensemble model based on adaptive feature weighting and improved whale optimization algorithm (LGWOA-AFWStacking) to predict the tribological properties of small sample composite lithium-based grease. The tribological test selected ILs-WS2 and ILs-MoS2 as additives and used MFT-R4000 reciprocating friction and wear machine to investigate the tribological properties of lubricating grease. First, machine learning models with excellent performance were selected as the base learners. Second, the Lévy flight strategy and golden sine algorithm were introduced to improve the whale optimization algorithm (LGWOA). Finally, based on LGWOA and base learner performance, the model adjustment coefficient was optimized adaptively. The corresponding weights were assigned to base learners according to the prediction precision, goodness of fit, and adjustment coefficient of each base learner. Weighted summation was realized. The experimental results demonstrated LGWOA-AFWStacking model could effectively predict the frictional properties of composite lithium-based grease, with predicted R2 values of 0.972 and 0.914 for average friction coefficient and wear width, respectively.