Online Robust RVFLN for Quality Prediction of Blast Furnace Ironmaking

In this paper, a new robust online regression prediction algorithm based on random vector functional-link network (RVFLN) is proposed for multivariate molten iron quality indexes modeling in blast furnace ironmaking process. Firstly, aiming at the multicollinearity problem of blast furnace ironmaking process data, the output weight of RVFLN is calculated by partial least squares regression. Secondly, in the actual ironmaking process, there are a wide range of abnormal values due to various reasons. Considering that these outliers exist in the direction of input and output samples at the same time, Mallows Generalized M-estimation is used to improve the robustness of the model. In addition, in order to realize online modeling, this paper combines online learning to improve the adaptability of the algorithm. Finally, the proposed online robust modeling algorithm is tested in the actual blast furnace ironmaking process data. The results show that this method has strong robustness and high prediction accuracy for abnormal data modeling.