An Efficient and Robust Saturation Pressure Calculation Algorithm for Petroleum Reservoir Fluids Using Neural Network

Saturation pressure is one of the key parameters in hydrocarbon reservoir engineering computations such as material balance and reservoir simulation. Empirical correlations (explicit methods) or equation of states (iterative methods ) are used to predict the fluid propertis. the accuracy of the mentioned computations will depend on the accuracy of the applied method. one of the greatest issues in calculation of saturation pressure , e.g.bubble poing pressure (Pb) , using EOS and iterative methods , is initial value to start the iteration . In thies work a feed forward multilayer neural network model is introduced to predict an initial value for bubble-point pressure in order to start the iterative methods. the model was developed using 411 published data from middle east and canada fields. 76 percent of data was used to train network 10 percent to cross validate of developed relationship during training process, 14 percent to test and trend analysis of the model. the results show that the model predicts a bubble -point pressure very close to exact one which can be used as an initial value in iterative methods . the proposed model provides prediction of bubble -point with relative average error of 0.532% absolute average error of 3.273%, a standard deviation of 3.417% and correlation coefficient of 0.999989.

Keywords:

saturation pressure , petroleum fluids , equation of state and neural network