Stochastic simulation of coalbed methane reservoir by radial basis function neural network

Abstract：Stochastic simulation method, known as Estimation ＆ Simulation Error（ ESE）, was ameliorated for coalbed methane（CBM） reservoir characterization. In the method, the unconditional simulation process was a vital step, which was achieved by radial basis function neural network（RBFNN） based on MATLAB and ArcGIS geographic information system （GIS）. The method was applied to generate many Stochastic simulation realizations of coal bed attributes rele- vant to CBM reserves from coalfield Geodatabase, including coalbed thickness, ash and moisture content, in the case study in Zhina coalfield, Guizhou, China. Two groups of gas reserves were calculated with the data obtained from eight- y-four wells by using fifty and one hundred of random realizations. Analysis of these reserves indicates that they have the analogical normal probability distribution ,the reserves range are 14. 972×10^8-16. 964×10^8 m3 and 14. 972×10^8- 17. 047×10^8 m3 respectively,and the average reserves are 15.92×10^8 and 15.97×10^8 m3 respectively. Three probabi- listic reserves, named as Pg0, P50 and P10, were acquired from the reserves by using one hundred of random realiza- tions. The probabilities of the real reserves more than P90（ 15. 389×10^8 m3 ） and P10（ 16. 611 ×10^8 m3 ） is 90% and 10% respectively. Methodology was presented for the cell-scale spatial variability of gas reserves probabilities, which was subsequently applied to the case study. The spatial variability of gas reserves probabilities is performed by the probabilities calculation in the fifty raster data layers of gas reserves from fifty random realizations.