A neurosimulation tool for predicting performance in enhanced coalbed methane and CO₂ sequestration projects

A tool for the practicing engineer to predict the important performance indicators that are critical in CO₂ storage projects in coal seams was developed. PSU-COALCOMP, a compositional coalbed methane reservoir simulator (hard computing protocol), was used to generate the necessary training data sets used in the training of the artifical neural networks (soft computing protocol) were developed. Reservoir porosity, thickness, permeability and permeability anisotropy ratio, sorption time constant, Langmuir volume and pressure constants for CH₄ and CO₂, relative permeability characteristics, temperature, and initial pressure and water saturation were the coal parameters that had important roles in determining the outcome of the project. The tested neural network predictions were accurate and sufficiently precise to establish confidence in the tool. Accordingly, the developed neural network could be used to screen thousands of possible scenarios of operational conditions in the optimization of the coal sequestration project design parameters in few seconds without going through intensive numerical simulations. The structure of the network is capable of establishing correct connections among the input and output neurons. The advantages of the proposed tool over the conventional reservoir simulators were discussed. This is an abstract of a paper presented at the 2005 SPE Annual Technical Conference and Exhibition (Dallas, TX 10/9-12/2005).