Prediction of water inflow from fault by particle swarm optimization-based modified grey models

Water inflow from fault (WIF) and its secondary impacts are the main environmental challenges in the mining industry. Traditional prediction methods for WIF are exceedingly challenging and costly. In this article, two exponentially weighted moving average (EWMA) modified grey models (GMs, i.e., EGM and REGM) were established to predict the WIF. Particle swarm optimization (PSO) algorithm was employed to optimize parameters of the models. Based on actual WIF data from Buliangou coal mine, the optimized models (i.e., EGM-PSO, REGM-PSO) were used to obtain the prediction equations for WIF. To investigate the validity of the proposed models, the differences between actual values and predicted values were analyzed, and comparison results were obtained by the commonly used GM and GM-PSO. Results show that, for the sample with the larger initial particle swarm and smaller inertia weight, there is a faster convergence speed of the PSO algorithm. Particle search efficiency in the PSO-optimized EWMA-GM is higher than that in the GM-PSO. Through the predicted results of WIF, it is found that the REGM-PSO is the best choice for WIF prediction, and the more historical information, the higher the predicted accuracy. Besides, the parameter optimization by the PSO, the EWMA optimization method and optimization of residuals all can improve the predicted accuracy. Predicted results also show that WIF will have a substantial growth in the future. Therefore, reasonable measures (e.g., draining and grouting) need to be taken to mitigate the damage caused by fault water inflow.