Perpustakaan Digital ITB

Production decline analysis is an important aspect of reservoir studies between the middle and the end of their producing lives, as it has a direct impact on the project's future development planning and economic analysis. The use of Arps decline curve analysis models (DCA) has been the standard method to forecast future production rates in the industry, this is due to its simplicity and reliability. However, this traditional method contains high subjectivity and requires a certain level of expertise to successfully model the production decline curve. It also requires a large amount of manual labor because the method is based on trial and error, this will take longer periods for fields with large numbers of production wells. Data-driven approaches that rely on machine learning algorithms have consistently produced good predictions in the last several years with various kinds of oil and gas industry applications, including time-series data. This could be an alternative solution to the problem. In this study, a Recurrent Neural Network (RNN) based algorithm, the Long Short-Term Memory (RNNLSTM) model will be developed using several production variables such as gas production rate, water production rate, and wellhead pressure as well constraint. This model will be used to forecast the future 180 days of oil production rate based on a set of field historical production time-series data. The forecast result will be then compared with Arps DCA and the actual production data to study both model's performance. Overall, the RNN-LSTM model performed well in comparison with the Arps DCA model, indicating this method might be applied in forecasting production rates and other variables that follow a similar pattern for other field cases.