Serious heterogeneity of coalbed, specific production process and complex laws of porosity-permeability changes during the producing period make mathematics models with superfluous assumptions very difficult to accurately describe coal reservoir so that they are not suitable for matching coalbed methane (CBM) well history gas production and forecasting it. As a new intelligent technique, BP neural network can be used for deal with nonlinear, non-stationary and complicated system problems, and it can quantitatively predict the short-term changes. Taking the case of a CBM well located in the south of Qinshui Basin which is the biggest production base of CBM in China, BP neural model for CBM productivity based on different temporal resolutions was established in this paper, and was applied to history match and quantitative prediction for gas production of CBM wells. The results show BP neural model can successfully match the production data and quantitatively predict it in different temporal resolutions. Meanwhile, the lower the temporal resolution is, the higher fitting and predicting accuracy the model has; while the higher the temporal resolution is, the clearer the model describes the changes of productivity in details.
展开▼
