Gas Compressibility Factor Prediction Using Machining Learning Algorithmic Protocol for Niger – Delta Gas Reservoir

Abstract

The gas compressibility factor also known as Z-factor plays an important role in obtaining thermodynamic properties of natural gas reservoir fluid property. Typically, empirical correlations and complex equation of state have been applied to determine this parameter in absence of laboratory measurements. However, high cost of running experimental measurement, poor performance and some limitations associated with these existing correlations have made the researchers to use intelligent models instead. Therefore, this study aimed at adopting support vector machine algorithm to forecast gas compressibility factor and to validate its performance with predictions from Artificial Neural network (ANN) and some existing correlations using statistical and performance plot analysis. A total of 519 data sets from Niger Delta was used in developing the model, out of it, 70 percent was used for training, 20 percent for testing and 10 percent for validation using MATLAB tool. From the statistical analysis result, it was observed that the new developed model did better than other existing methods with numerical value of 0.1997 rank, 0.0009 mean absolute error and 0.98 of coefficient of correlation using the test data. The cross plot of the support vector machine model gave the tightest cloud along 45o reference line. The residual (error) associated with the performance was impressive which was done to observe the distribution and the interval at which the error is minimal.