Production Prediction Based on Principal Component Analysis and Neural Network—A Case Study in a Coalbed Methane Field

Great differences in well gas productions and complicated geological influential factors are the challenges to make rational development strategy for a coalbed methane field located in the Bowen basin of southeast Queensland, Australia. The geological influential factors such as measure depth, coal thickness, coal density, gas content, max vitrinite reflectance, permeability, dip, curvature, overlaying sandstone thickness, distance to fault are analyzed with principal component analysis and a new set of uncorrelated components is achieved. Then the strong interaction which may exist between these components and well productions are identified by neural network, and the production prediction model is generated using the established internal relationship. The results show that the main principal components extracted by principal component analysis reflect reservoir storage ability, well production ability and well drilling engineering ability. Three types of production region are classified from the prediction model, and type I in the western edge is favorite area with high-yielding production and type II in the central region is the sub-favorable production area. The good production performances of a new drilled well in the high-yielding area validate the effectiveness of the production prediction model.