Log interpretation for lithology and fluid identification using deep neural network combined with MAHAKIL in a tight sandstone reservoir

As an important source of unconventional natural gas, tight sandstone reservoirs have attracted considerable attention. The accurate identification of lithological and fluid-bearing zones is a limiting factor for exploring such reservoirs effectively. Log data, which provide abundant and high precision geological information, are always used for this task. Here, we combined a deep neural network with an oversampling approach named MAHAKIL for lithology and fluid identification using log data from a tight sandstone gas reservoir. MAHAKIL was adopted to solve the class imbalance problem resulting from imbalanced training samples, and the outputs were then fed into a deep neural network to learn the complex and abstract geological patterns related to the lithology and fluid in a layer-by-layer manner. We first demonstrated the unsatisfactory performance of a typical classification algorithm on a simulated data which conforms to the normal distribution and is imbalanced. Then, the performance of our method was compared with that of a support vector machine and a deep neural network on the imbalanced actual data, and the F-beta score which is the weighted harmonic mean of precision and recall was used as the evaluation criterion. The results show that the F-beta score of our method was generally higher than that of the other two methods, which indicates the superiority of the proposed method for lithology and fluid identification of tight sandstone gas reservoirs when the learning samples are imbalanced.