Study on logging interpretation of coal-bed methane content based on deep learning

To solve quantitative interpretation problems in coal-bed methane logging, deep learning is introduced in this study. Coal-bed methane logging data and laboratory results are used to establish a deep belief network (DBN) to compute coal-bed methane content. Network parameter effects on calculations are examined. The calculations of DBN, statistical probabilistic method and Langmuir equation are compared. Results show that, first, the precision and speed of DBN calculation should determine the restricted Boltzmann machine’s quantity. Second, the hidden layer neuron quantity must align with calculation accuracy and stability. Third, the ReLU function is the best for logging data; the Sigmoid function and Linear function are second; and the Softmax function has no effect. Fourth, the cross-entropy function is superior to MSE function. Fifth, RBMs make DBN more accuracy than BPNN. Furthermore, DBN calculation accuracy and stability are better than those of statistical probabilistic method and Langmuir equation.