The application of artificial neural networks for the prediction of water quality of polluted aquifer

From hydrocarbon reservoirs, beside of oil and natural gas, the brine is al so produced as a waste material, which may be discharged at the surface or re-injected into the ground. When the wastewater is injected into the groun d, it may be mixed with fresh water source due to to several reasons. Forec asting the pollutant concentrations by knowing the historical data at sever al locations on a field has a great importance to take the necessary precau tions before the undesired situations are happened. The aim of this study is to describe Artificial Neural Network (ANN) approa ch that can be used to forecast the future pollutant concentrations and hyd raulic heads of a groundwater source. In order to check the validity of the approach, a hypothetical field data as a case study were produced by using groundwater simulator (MOC). Hydraulic heads and chlorine concentrations w ere obtained from groundwater simulations. ANN was trained by using the his torical data of last two years. The future chlorine concentrations and hydr aulic heads were estimated by applying both the long-term and the short-ter m ANN predictions. An approach to overcome the effects of using the data of a single well was proposed by favouring the use of data set for a neighbou r well. The higher errors for the long-term ANN predictions were obtained a t the observation wells, which were away from an injection well. In order t o minimise the difference between the results of long-term ANN approach and flow simulation runs; the short-term prediction was applied. The use of sh ort-term prediction for the wells away from an injection well was found to give highly acceptable results when the long-term prediction fails. The ave rage absolute error obtained from the shortterm forecasting study was 3.5% when compared to 18.5% for the long-term forecasting.