CAJ | 학술논문

多元线性回归及 BP 神经网络是煤层含气量测井解释的常用方法。基于澳大利亚 Galilee 盆地和沁水盆地煤层测井资料和实测含气量数据，通过相关性分析和显著性检验，筛选了和含气量相关的测井参数，通过多元线性回归建立含气量与测井参数的解释模型；基于 BP 神经网络的理论，通过网络训练和测试，建立了煤层含气量和测井参数的非线性解释模型。讨论了多元线性回归模型的参数选择方法，并对两种解释方法的误差特点进行了分析，讨论了两种方法的适用性。结果显示：多元线性回归法和 BP 神经网络法是煤层含气量解释的常用方法，前者的解释误差比后者大；多元线性回归法解释精度与煤层含气量相关，适用于含气量较高的井；BP 神经网络法解释精度普遍较高，在含气量高和低的井中均可适用，解释效果受输入层样本的数量和质量影响，样本数量越多，区域代表性越强，解释效果越好。
다원선성회귀급 BP 신경망락시매층함기량측정해석적상용방법。기우오대리아 Galilee 분지화심수분지매층측정자료화실측함기량수거，통과상관성분석화현저성검험，사선료화함기량상관적측정삼수，통과다원선성회귀건립함기량여측정삼수적해석모형；기우 BP 신경망락적이론，통과망락훈련화측시，건립료매층함기량화측정삼수적비선성해석모형。토론료다원선성회귀모형적삼수선택방법，병대량충해석방법적오차특점진행료분석，토론료량충방법적괄용성。결과현시：다원선성회귀법화 BP 신경망락법시매층함기량해석적상용방법，전자적해석오차비후자대；다원선성회귀법해석정도여매층함기량상관，괄용우함기량교고적정；BP 신경망락법해석정도보편교고，재함기량고화저적정중균가괄용，해석효과수수입층양본적수량화질량영향，양본수량월다，구역대표성월강，해석효과월호。
Multiple linear regression and BP neural network are gas content logging interpretation methodologies commonly used in coal seam. Based on well logging data and measured gas content of CBM well in Galilee basin of Australia and Qinshui basin of China, this study screened the logging related parameters of gas content through correlation analysis and then established the relationship model between gas content and logging parameters. Based on BP neural network theory, this study not only established a nonlinear prediction model of CBM gas content and logging parameters through the network training and prediction, but also analyzed the error of the two methods and discussed their applicability.