Data-driven modeling for online fault pre-warning in thermal power plant using incremental Gaussian Mixture Regression

Abstract

This study introduces a data-driven model for online fault pre-warning in thermal power plant using incremental Gaussian mixture regression. To tackle the issue of outdated parameters in existing fault pre-warning models, this study puts forth an incremental Gaussian mixture regression that leverages the merging of Gaussian components to reconstruct the model and enable online modeling. Due to its criticality, a forgetting factor is introduced during the merging process to efficiently manage the weight allocation between present and historical patterns, thereby guaranteeing the model’s accuracy. The results of the sine function case demonstrate that the IGMR model exhibits excellent pattern control performance and modeling efficiency. Furthermore, the IGMR model is employed to forecast parameter alterations in pulverizer blockages with mode switching, and experimental validation indicates that IGMR precisely anticipates parameter changes following mode switching. Compared to on-site solutions, the pre-warning of coal blockage has a clear advantage in advance warning.