Perpustakaan Digital ITB

As the world is on the brink of entering the fourth industrial revolution, digitalization will play a fundamental role in almost every sectors. In the last few decades, digitalization and artificial intelligence have been incorporated to almost every industry in order to increase efficiency. Oil and gas, as the biggest suppliers of affordable energy, has surely started adapting the technology in every sector of its’ industry. In line with the previous statement, this research discusses the application of Time Series for production rate forecasting. Decline Curve Analysis (DCA) is currently the most popular and reliable approach for production rate forecasting, as it is an empirical reservoir engineering technique that extrapolate trend from oil and gas wells. The objective of this research is to provide an alternative approach of production rate forecasting by using machine learning (ML). A machine learning-augmented systematic workflow is then constructed to forecast production rate from a given well rate data. Time Series is used to decipher production rate data from a certain time span and then forecast rate for the upcoming time. The forecasted rate will be generated with no specific decline curve type (e.g. exponential, hyperbolic, harmonic) as the one described by J.J. Arps. Production rate forecasting is fully conducted based on the trend data learnt by the Time Series model. Therefore, there will be no such term as ‘optimist’ or ‘pessimist’ forecast. A thorough analysis and data cleaning was conducted to the production rate data. Then, a specific segment out of the whole production rate data was selected for the forecasting purpose. This selected segment was then used to train the time series model. The trained model was utilized to generate production rate in a user-specified upcoming time. Comparison with the result obtained from commercial DCA software showed that the Time Series model can forecast rate with accuracy more than 94%. The forecasted rate is then eligible to be used for further purposes, such as artificial lift design and selection. This research will provide oil and gas practitioners a novel approach of forecasting production rate which can reduce cost, increase efficiency, and has infinite potential to be further developed.