Machine learning based decline curve analysis for short-term oil production forecast

Traditional decline curve analyses (DCAs), both deterministic and probabilistic, use specific models to fit production data for forecasting. Various have been applied unconventional wells, including the Arps model, stretched exponential Duong combined capacitance-resistance model. However, it is not straightforward determine which model should be used, as multiple may a dataset equally well but provide different forecasts, hastily selecting probabilistic DCA can underestimate uncertainty in forecast. Data science, machine learning, artificial intelligence are revolutionizing oil gas industry by utilizing computing power more effectively efficiently. We propose data-driven approach this paper performing short term predictions production. Two states of art level tested: DeepAR used Prophet time series analysis on petroleum data. Compared with traditional using models, learning regarded as” model-free” (non-parametric) because pre-determination required. The main goal work develop apply neural networks techniques without having substantial knowledge regarding extraction process or physical relationship between geological dynamic parameters. For evaluation verification purpose, proposed method selected Midland fields from USA. By comparing our results, we infer that useful gaining better understanding behavior mitigate over/underestimates resulting single In addition, performs spreading forecasting; is, end up forecast outperforms standard methods.