Publications

Measurement of coal seam properties, like measurement of conventional reservoir properties, always entails some errors. To understand the impact of these errors on the Warrior and the San Juan Basin coal well performance Zuber and Olszewski employed different model datasets of Sawyer et al. to investigate the influence of a 1% change in key reservoir properties and for Monte Carlo simulations. This paper presents the utilization of Artificial Neural Networks (ANN) for development of a proxy model in reservoir simulation studies to be used in uncertainty analysis on CBM reservoirs. Several simulation runs were generated for training, validation, and testing of these neural networks in order to predict the drainage efficiency of gas. Uncertainty in reservoir properties is taken into account by varying the reservoir parameters within an estimated range of value. The overall performance levels are optimized by varying network parameters, monitoring the error values for testing datasets, and the descending order of the most influencing parameters. The parametric study of 1% in variation in a variable (while holding all other variables constant) showed gas drainage efficiency to be most sensitive to, in order of descending importance, besides the well spacing, fracture permeability, initial water saturation, initial reservoir pressure, fracture porosity, and pressure constant. The uncertainty analysis was employed with 1% variation of all input parameters for 30-years. A small number realization of reservoirs are required to develop the neural network but thousands of simulation runs can be made in a matter of seconds. Keywords: Simulation Study, Optimum Well Spacing, Artificial Neural Network, Uncertainty Analysis