Construction and application of prediction methods for coal texture of CBM reservoirs at the block scale

Abstract

Identifying coal texture has an important influence on the exploration and development of coalbed methane (CBM), but the prediction at the block scale is still insufficient. This paper combines the core idea of machine learning algorithm, coal body deformation mechanism, and critical layer theory to establish a prediction method for coal texture of CBM reservoir at block scale. The factor analysis method was used to eliminate the information redundancy of logging data, and the key logging parameters for identifying the coal texture were optimized. These include density (DEN), natural gamma ray (GR), X well diameter (CALX), and deep lateral resistivity (LLD). The K-means clustering algorithm was used to extract the numerical characteristics of crucial parameters. The K-nearest neighbor (KNN) classification algorithm established the identification equation of coal texture. The following classification was obtained: primary structural coal (T ≥ 0.70), broken-mylonite coal (T ≤ −0.50), and cataclastic coal (−0.50 < T < 0.70); these were used to accurately and finely divide the coal texture of multiple coal seams. Subsequently, from the fact that the structure of the coal body of a coal seam is necessarily influenced by the critical layers of the coal-bearing rock system in which it is located, the value of S for the strength of the coal-bearing rock system is proposed. According to the deformation characteristics of the coal body and the critical layer theory of the coal reservoir in the mining area, the length of the coal-bearing rock system section is determined as 30 m for the top plate and 20 m for the bottom plate of the coal seam and combined with the strength adjustment coefficient. The S value calculation is derived. The S value was calibrated using the calculation results of the coal texture logging identification model: primary structural coal (S ≥ 4.50), cataclastic coal (3.50 < S < 4.50), and broken-mylonite coal (S ≤ 3.5). Finally, the above methods are used to identify the block scale coal texture of the main coal seam of Kele syncline in western Guizhou. The identification prediction results are the same as those of underground sampling.