Bottom-Hole Pressure Prediction from Wellhead Data Using Developed Machine Learning Models

Abstract

The accurate prediction or measurement of bottom-hole pressures in
oil and gas reservoirs cannot be over-emphasized in the Petroleum
Industry. Mechanistic, numerical and analytical models have been
developed and deployed to determine bottom-hole pressure. Some of
these models developed have failed to predict bottom-hole pressures
to an acceptable accuracy. However, the down-hole gauges measure
the bottom-hole pressures to an acceptable accuracy, but, are
expensive to use and maintain. This study focused on developing
models using random forest regression and gradient boosting
regression to predict bottom-hole pressures in oil and gas reservoirs
accurately. The input data used was collected from the Volve Field
(Jurassic sandstone reservoir) and filtered and correlated
successively. The data was normalized using Python programming
to prepare the data sets for input into the model. The results showed
that the random forest regression model has an accuracy of 97.80%
while the gradient boosting regression model has an accuracy of
95.83%. The average magnitudes of the errors are 0.0067 and
0.01266 for random forest and gradient boosting regression models
respectively. The developed models predicted the bottom-hole
pressures for the reservoir with an acceptable degree of accuracy
and error magnitude. The Random Forest Regression and the
Gradient Boosting Regression models were seen to be economical
and accurate in solving the problem of predicting bottom-hole
pressures in oil and gas reservoirs