Perpustakaan Digital ITB

Liquid loading in gas production wells is a serious production problem. Liquid loading can reduce production capacity and even kill the well at any time. Therefore the prediction of the critical rate of liquid loading must be accurate. An accurate critical rate prediction will also provide an accurate prediction of the time liquid loading occurs in a well. Currently a popular method for calculating the critical rate of liquid loading of a well is to use empirical equations such as the turner equation, Coleman Equation and Li Min Equation. The popular method used today has several drawbacks in calculating Interfacial Tension, which can give a large difference between the predicted critical rate and the actual critical rate. This study uses a machine learning approach to predict the critical rate of liquid loading with limited data. The algorithms used in this study are KNN, Gradient Boosting, Extreme Gradient Boosting and Random Forest. The dataset was taken from 49 wells that had experienced liquid loading in field Y with swamp zone characteristics. The most suitable machine learning model will be used to predict the critical rates of two wells that have never experienced liquid loading. Decline Curve Analysis is used to predict when these two wells will experience liquid loading. The results of the analysis of this study show that the best machine learning model for calculating the critical rate of liquid loading in field Y is the random forest. This is because the random forest has good performance evaluation parameter values and has the lowest possible overfitting among other machine learning models. The Random Forest model is used to calculate the critical rate of the Y-TM1 and Y-TM2 wells. Using Decline Curve Analysis liquid loading will occur in well Y-TM1 on September 2023 and well Y-TM2 on late July 2023.