A sequential structure for water inflow forecasting in coal mines integrating feature selection and multi-objective optimization

Abstract

To construct an accurate and stable approach for water inflow forecasting, a series of advanced and effective techniques, such as variational mode decomposition (VMD), outlier robust extreme learning machine (ORELM) and multi-objective grey wolf optimizer (MOGWO), are appropriately integrated into this study. Considering that the influence of the mode number on theVMDdecomposition effectiveness, such an argument is determined by observing the converged centre frequency distribution among the components. Then the characteristic items of water inflow series are extracted by VMD, thus obtaining a series of sub- components. Afterwards, ORELM is applied to predict each component, where the parameters of ORELM are optimized by MOGWO with multi-objective functions including forecasting accuracy and stability. Correspondingly, the aggregation of all components' prediction values is considered as the final results. The experimental results obtained by performing eight various models on real-time data demonstrate that the supplementary modules achieve positive effects on the improvement of prediction accuracy, where the proposed model implements an average performance promotion of 48:43% compared with the contrastive models.