Perpustakaan Digital ITB

The oil and gas industry has seen numerous advancements in recent years, which have impacted regional and global businesses and economies associated with the oil industry, as well as other energy sectors. The notion of machine learning, also known as ML, is an application of Al that is focused with the improvement of programs that can access all past data directly and indirectly and learn from various market and operational experiences. Machine Learning in gas lift can provide significant economic value since they enable one to simulate the behavior of a real or planned gas-lifted well using a digital replica of the well. Database for the machine learning is made based on two actual well data that being generated using random number and standard deviation value. The database will be simulated one by one using the prosper software to get the max oil rate, injected gas rate and injection pressure output. Then the data goes into preprocessing to determine clustering method that being used, which is k-means. To ensure the prediction value of this machine learning, three regression method are used (Linear Regression, Gradient Boosting, and Ridge Regression). These three methods will be compared by the resulting r-squared value. The method that produces an r-squared value close to 1 is the choosen method. Based on the result, the best regression method is linear regression method with the highest r-squared value on the parameters of max oil rate and parameter injection pressure. R-squared value for max oil rate parameter is 0.989 on the first set data and 0.992 on the second set data. For injection pressure parameter, the r-squared value is 0.999 on the first set data and 0.843 on the second set data. However, for the injected gas rate parameter, it is difficult to choose the method because the r-squared value obtained is very low. Therefore, to determine the injected gas rate, the decision was made to use analytical equation instead of machine learning.