Prediction of Bottom-hole Pressure in Oman oil fields using AI Techniques

The flowing bottom-hole pressure (FBHP) of an oil well is important information for reservoir engineers and production technologies. It is an essential parameter to enable well production forecasting, production monitoring and well artificial lift system optimization. Therefore, the installation of down-hole gauges in oil wells has become a common practice in the oil petroleum industry; especially in wells lifted with electrical submersible pumps. However, these down-hole gauges require continuous maintenance and calibration. Also, due to the harsh down-hole environment there is a high risk they will fail and will be expensive to repair or replace. In addition, intervening a well from time to time to measure the FBHP is an expensive task and associated with production risk and interruption. For these reasons the motivation of the estimation of the FBHP has come out. The objective of this study is to develop an artificial techniques models to predict the FBHP in vertical oil wells by using measured data from oil fields of Oman. This can be achieved by designing and developing a single and multilayer perceptrons feed-forward neural network (FFNN) with back-propagation algorithm. The selection of the best number of hidden layers can be done through trial and performance assessment. Similarly, the best number of neurons in each hidden layer will be selected. Also, a radial base neural network model will be developed and compared to the single and two hidden layers models performance, neuro-fuzzy can be addressed as well. Real field data is used to train and test the model. Field data has been gathered from pumped oil wells from Petroleum Development of Oman company oil fields. Model inputs data were selected from list of available surface well measurements. Data is preprocessed prior to training and testing processes. The model performance evaluation conducted by means of testing the model against actual well down-hole measurements data from the same oil fields. These testing data have not been used in the model training. Standard statistical analyses preformed on the results to evaluate the models prediction accuracy such as; root mean square error, standard deviation of error, minimum/maximum/average absolute relative error and correlation coefficient.