关键词: 支持向量机, 人工神经网络, 多元回归分析, 参数乘积判别法, 致密砂岩, 含气性评价

Abstract: Four different methods, including support vector machine (SVM), artificial neural network (ANN), multiple regression analysis (MRA) and parameter product decision (PPD), were applied to the gassiness evaluation of forty gas-bearing layers in the tight sandstones of Tabamiao Area in Ordos Basin. Their mean absolute relative residual values between predicated results and gas test results are 0, 4.63%, 29.71% and 18.75%, respectively. This case study shows that the former two nonlinear methods (SVM, ANN) are very superior to the later two linear methods (MRA, PPD). And the SVM is superior to ANN, while PPD is in turn superior to MRA. That is because there exists a complex and nonlinear relationship between gassiness and its related geological factors such as porosity, permeability and gas saturation. Therefore, ANN and SVM should be adopted to describe any complex relationship between the study target and its related multi-geological-factors. In particular, for time-consuming tasks of data processing with repetitious multi-geological-factor analysis, it is recommended that SVM should be used, because it is much faster and more precise than ANN. On the other hand, the case study also indicates that PPD has its advantages of conciseness and high speed. PPD has more precision than MRA, while MRA is fast in processing speed and can be used as an auxiliary tool to establish the order of dependence between the target and its related multi-geological-factors which cannot be estimated using the other three methods.

Key words: support vector machine, artificial neural network, multiple regression analysis, parameter product decision, tight sandstones, gassiness evaluation