Development of artificial neural network of mine dewatering
Open pit mines often experience problems related to groundwater inflows. To perform mineral extraction in safe conditions with high productivity, it is essential to have dry working conditions. For this reason, the groundwater table is often lowered below the elevation of the floors of the pits by using various dewatering schemes. Numerical groundwater models are powerful tools that can be used to simulate the behaviour of aquifers during dewatering operations. However, these models typically require a lot of geohydrological data which are often expensive and time-consuming to collect. When geohydrological input data are limited, artificial neural networks (ANNs) provide an alternative way of predicting the behaviour of the groundwater system under dewatering conditions. ANNs can simulate complex systems, and have been used to provide simple and accurate solutions to problems encountered in many disciplines of the earth sciences.This study investigated the possibility of predicting the impacts of pit dewatering on the aquifer system in the vicinity of open pit mines where geohydrological inputs are limited, using ANNs. First, the performance of the ANNs in predicting hydraulic head responses was evaluated by using synthetic datasets generated by a numerical groundwater model developed for a fictional mine. The synthetic datasets were then used to both train and evaluate the performance of the ANNs. The ANN found to give the best predictions of the hydraulic heads had an architecture of 2-6-1 (input-hidden-output layers) and was based on the hyperbolic tangent transfer function. This network was selected for application to real open pit mines. The selected ANN was next used to predict hydraulic heads at a number of piezometers installed at two open pit mines in the Democratic Republic of the Congo. The only input to the ANN was the recorded hydraulics heads and the time of recording. A portion of the real dataset was used to train the ANN, while the remaining portion was used to evaluate the performance of the ANN in predicting the hydraulic heads. The results of the performance analyses indicated that the ANN successfully predicted the general behaviour of the aquifer system under dewatering conditions, using only limited input data. The results of this investigation illustrate the great potential of using ANNs to predict aquifer responses during dewatering operations in the absence of comprehensive geohydrological datasets. Since these networks recognise patterns in the training datasets without considering the underlying physical principles that govern the processes, the responses of complex systems that are dependent on numerous parameters may be predicted.