Accuracy Assessment of Single and Hybrid Models for Predicting Shale Gas Production

Abstract

The accuracy of a predictive tool determines the levels of trust in the model and its attraction for commercial usage. The study examined the single and hybrid model approach for shale gas production. Multilayer perception artificial neural network (ANN), autoregressive integrated moving average (ARIMA), and Arps–power law exponential hybrid decline models were developed to predict shale gas production and compared with the already developed Arps decline and power law exponential (PLE) decline models. By a trial-and-error approach, a multilayer perception (MLP) network with four neurons in the hidden layer was attained in the ANN structure to predict shale gas production. While for the ARIMA model, the number of nodes that showed the best performance indicated (2,1,2) for the two sets of data. Evaluation of the root mean square error (RMSE) values for the models showed that the Arps–power law exponential hybrid decline model had a lower percentage error in conjunction with good accuracy. The study found the Arps–power law exponential hybrid decline model to be a good forecaster of shale gas production and that hybrid models do deliver better accuracy over single models. A future revision of model assumptions may improve its accuracy and make the Arps–power law exponential hybrid decline model an attractive predictive tool.