Reconstruction and prediction of capillary pressure curve based on Particle Swarm Optimization-Back Propagation Neural Network method

Capillary pressure curve plays a critical role in the reservoir evaluation. It is essential to reconstruct and predict capillary pressure curve properly. Many traditional capillary pressure correlations have been suggested in the literature. However, their major limitation is mainly applicable to homogenous reservoir, and the larger error will be caused when heterogeneous reservoir is dealt by using these mathematical correlations. This study aims at providing an important method based on Particle Swarm Optimization-Back Propagation Neural Network (PSO-BP neural network) to represent and predict capillary pressure curve for homogenous and heterogeneous reservoir. The combination of PSO algorithm and BP neural network converges quickly, which improves the accuracy and efficiency of simulation. In this paper, core samples from three blocks of the same marine-sand reservoir, whose porosity is between 0.6% and 20.0% and permeability is between 0.1mD and 6117mD, are investigated by PSO-BP neural network method and J-Function method respectively. The reconstruction and prediction results are compared with the results obtained by mercury intrusion method in laboratory. The results show that capillary pressure curves reconstructed and predicted by PSO-BP neural network method are in better agreement with mercury intrusion curves than J-Function method, with 0.1%-5% and 5%-8% relative error respectively, which can totally meet the in-situ requirements. It is also demonstrated that PSO-BP neural network method is more suitable for homogenous and heterogeneous reservoir.