Carbonate Seismic Geopressure Prediction Using Neural Networks

The prediction of high-pressure zones in carbonate reservoirs presents a major challenge in reservoir geophysics, as we cannot explore the assumption as in clastic reservoir prediction that the pressure mechanism is due to mechanical compaction. In this paper we use unsupervised and supervised neural network methods to predict geopressure in carbonate reservoirs. We test two approaches. The first approach uses multilayer perceptrons method, a supervised neural network, to predict porosity and in turn uses self-organizing map (SOM), an unsupervised neural network method, to predict geopressure from porosity. The second approach also uses SOM method but use both seismic impedance and estimated porosity together to predict pore pressure. We use a real data example to demonstrate that these approaches are robust in clustering the seismic attributes from a carbonate reservoir.