博文盆地M煤层气田含气量主控因素分析及预测

精准预测煤层含气量是煤层气资源量评价及开发方案制定的关键之一，传统预测方法应用范围有限且预测误差较大。为了进一步提高含气量预测精度，以博文盆地M煤层气田GM煤层为研究对象，基于皮尔逊相关分析法对含气量主控因素进行定量分析，优选出了9个煤层含气量影响主控因素，提出基于高斯过程回归、支持向量机及集成装袋树等机器学习算法的煤层含气量预测模型，对比分析各方法优缺点，进而优选出最佳的含气量预测模型。研究结果表明：高斯过程回归精确度最高，预测结果平均相对误差为4.5%,适用于本气田煤层气的含气量预测；支持向量机泛化能力强，但是超平面的绝对化导致预测精度降低；集成装袋树并不侧重于训练数据集中的任何特定实例，对于噪声数据其基本不受过分拟合的影响。

Analysis and Prediction of Main Controlling Factors of Gas Content in Coalbed of M Coalfield in Bowen Basin

Accurate prediction of coalbed gas content is one of the keys to the evaluation of coalbed gas resources and the determination of development plans. Traditional prediction methods have limited application scope and big prediction errors. In order to further improve the prediction accuracy of gas content, taking the GM coalbed of the M coalfield in Bowen basin as the research object, the main controlling factors of gas content were quantitatively analyzed based on the Pearson correlation analysis method, and nine main controlling factors affecting the gas content of coalbed were selected. The prediction model of coalbed gas content based on machine learning algorithms such as Gaussian Process Regression, Support Vector Machine and Ensemble Bagging Tree, etc., were proposed. After comparing and analyzing the advantages, disadvantages of each method, the best prediction model for gas content was selected. The research results show that the Gaussian Process Regression has the highest accuracy, and the average relative error of the prediction results is 4.5%, which is suitable for the gas content prediction of coalbed gas in M coalfield. The Support Vector Machine has strong generalization ability, but the absoluteization of the hyperplane reduces the prediction accuracy. The Ensemble Bagging Tree does not focus on any particular instance in the training dataset and is largely immune to overfitting on noisy data.