Application of Artificial Neural Network to Evaluate Coalbed Methane Exploration Potential: A Case Study from Permian Longtan Formation, Shuicheng, Guizhou

The Shuicheng area in Guizhou (China) contains extensive coalbed methane (CBM) resources in the Permian Longtan Formation and it has high potential for large-scale exploration and development. However, exploration and development have started in a few areas in these regions. Evaluating CBM exploration potential (EP) areas is crucial for efficient exploration. Using an artificial neural network (ANN) model, this research established a CBM EP model for evaluating prospective areas in the Dahebian, Shenxianpo, and Tudiya CBM blocks. Thirty-four coal seams exist in the research area, with coal seams 01(#) and 07(#) highly developed exclusively in Dahebian and Shenxianpo blocks, coal seams 09(#), 11(#), 12(#), 13(#), and 14(#) throughout the three blocks, and coal seam 25(#) in the southern portion of Tudiya blocks. The coal seam 11(#) is the thickest (0.2-11.48 m), with buried depth of 102-1522 m, and is characterized by a wide range of properties variation: low- to high-rank coal (Ro% 0.6-2.5), gas content range of 0.43-20.12 m(3)/t, and weak to highly deformed by a string of normal and reverse faults. Considering six key evaluation parameters (thickness, buried depth, gas content, rank, structural and roof sealing conditions), an EP model was developed, and the analytic hierarchy process membership function and pairwise comparison matrix were used to calculate the weight of each parameter. A mathematical model has determined the comprehensive CBM EP coefficient and the evaluation scores ranged 0-1; the high scores represent high CBM EP. In total, 101,965 data points of the six input neurons and one target neuron were extracted from the 3D parameter model and the result of the EP model, respectively, and were used to train the ANN model. The model performed high accuracy based on R and R-2 and minimum mean square error and root mean square error. The results reflect that the model construction was 99% accurate, and the established EP model in this study was also 99% precise and adequate for evaluating the favorable areas for exploration. The results show that the EP values of > 0.8 represents medium to thick coal seam, weakly deformed medium-rank coal having high gas content and buried within 600-1200 m. The EP values of 0.65-0.8 indicate medium thick, medium to high-rank coal with weak-moderately deformed coal having a depth range of 400-1200 m. In areas with EP values of < 0.65, coal seams are thin, highly deformed, middle to high-rank coal with medium to high gas content and < 400 and > 1200 m buried depth. Two very high-potential (VHP), three high-potential, and two medium-potential areas were identified; among them, VHP-1 in the Biandanwan area had the most significant potential, which is supported by CBM development data. This model can significantly impact successful CBM exploration and development of complex geological settings areas in the Guizhou province in the future.