Perpustakaan Digital ITB

Decline Curve Analysis (DCA) is a widely used method in the oil and gas industry for forecasting production rate by simply analyzing trends from the production history. However, traditional DCA model (Arps model) often overestimate production in unconventional reservoir due to their unique flow regimes. This study evaluates four alternative models beside Arps model as more accurate approach for forecasting in unconventional reservoir, particularly shale gas reservoirs. This study also investigates the impact of automatic decline curve analysis as it is becoming increasingly important and machine learning-based outlier detection on improving the prediction accuracy. Using the ‘curve_fit’ function from SciPy libraries in Python, the ADCA is implemented to find the best fitting parameters. Additionally, Orange software is used to implement Isolation Forest and Local Outlier Factor for the outlier detection. The results show that the alternative DCA models provide more accurate and reliable predictions than the Arps model. The results also show that by using the machine learning-based outlier detection significantly improves the predicted accuracy, minimizing the average Mean Absolute Percentage Error (MAPE) from 60.4% to 14.8% for Well 1 and 36.4% to 19.6% for Well 2.