Using artificial neural network to predict gas production for infill wells

Abstract

Nowadays, more and more fields all over the world are trying to improve gas recovery factor. Infill drilling is one effective method for this purpose. The expected gas production rate and cumulative gas production is a key component in determining whether or not to drill a well. The success of adding a new well to the field depends on the accuracy of prediction of the gas production, as the more accurate the prediction is, the better the decision on drilling location will be. One interesting technique to predict an appropriate location for infill well is Artificial Neural Network (ANN) which can learn from the historical data to create a representation of complex relationship between input and output samples. In this study, the ANN is applied in conjunction with numerical reservoir simulation. Production data generated from a numerical simulator were used to train the network to forecast gas production at undrilled location. Many input parameters were considered, screened, and chosen in order to study their effect on the result. A few case studies were performed to highlight the importance of these input parameters. Finally, the prediction performance of ANN was evaluated. The results show that the ANN can be effectively used to predict gas production with accurate prediction performance. However, substantial errors still occurred at some well locations due to the inaccuracies when using ANN to predict the output based on an extrapolation basis.