A Two-stage Heterogeneous Ensemble Learning Algorithm Based on MLP-SVR-LGBM

A single machine learning approaches often encounter numerous challenges and limitations when addressing complex problems. In this study, we propose a novel stacking ensemble learning model for heterogeneous soft sensors, utilizing multi-layer perceptron (MLP), support vector regression (SVR), and light gradient boosting machine (LGBM) as base learners, with gradient boosting decision tree (GBDT) as the meta-learner. First, the proposed heterogeneous soft sensor model fully considers the diversity of the base learners and their learning performance to effectively exploit the advantages of various machine learning methods. Second, the meta-learner, GBDT, is an integrated learning algorithm that can better handle different data of the base learners and improve the performance of the overall model. Subsequently, we demonstrate the superiority of the proposed method is confirmed through an artificial dataset case study.