Neural network aided design of oil production units

The paper presents a novel neural network based method for predicting the performance of the multi-stage gas/oil separation plants in crude oil production. The developed method can play an important role during planning and operation of oil production facilities. It predicts important oil properties, which are usually obtained from expensive and time-consuming laboratory tests. The prediction is based mainly on the basic oil composition analysis from standard lab tests. The neural networks were trained using data collected from laboratory tests. The objective of these tests is to find the best oil/gas separation stages to minimize the separator gas/oil ratio and to improve the resulting oil specific gravity. The neural networks accept the initial and final pressures and temperatures of each stage and then try to utilize the oil composition information to predict the stage gas/oil ratio. Two neural networks were built, one for the initial stage with difference of pressure over 400 psi, and the second for the separator stages covering the lower range of pressure. The method can also be useful simulation tool in optimizing the oil production operation.