Abstract

Coal lithology is predicted from back propagation neural network in Jharia coalfield. Three wells which are viable for coalbed methane exploration are chosen for this study. This work demonstrates the utilization of multilayered feedforward neural network for identification of coal litho- units such as: coal, shaly coal, carbonaceous shale and jhama. The input parameters from well logs for this network are bulk density, gamma ray, long normal resistivity and neutron porosity. Output parameter is litho- unit for coal, shaly coal, carbonaceous shale and jhama. Different litho-codes are assigned to these litho-units. The network is optimized with minimum sum-squared error of training and testing dataset for 150 epochs and 3 hidden nodes. The model predicted litho-codes for other two wells matches well with the observed litho-codes with R2 = 0.99.