Methods of forecasting explosive gases in mines based on dynamic neural networks

Abstract

Some methods of forcasting of explosive gases content in mine workings based on dynamic simulated neutral network were developed and implemented in order to improve safety operations in mines. The methods include a nonlinear autoregressive with exogenous inputs (NARX) and a Time-Lagged Feedforward Neural Network (TLFN). The selection of network architecture (determination of the number of neurons in the hidden layer and the length of the delay) was done basing on the on the minimum value of the mean square error. The adaptation of the parameters of the suggested networks was based on metaheuristic algorithms for clonal selection of two types: with the use of simulated annealing and without it. In order to evaluate the effectiveness of the suggested methods numerical studies, which prove the effectiveness of the selected networks applicable to the specific conditions of use, were done.