A quantitative evaluation model for biodegraded reservoirs based on multinomial logistic regression

The application of multiple evaluation technologies is a prerequisite for improving the probability of correctly identifying the type of a given reservoir (e.g., gas, oil, or biodegraded oil reservoirs). The traditional identification methods do not consider the graphical information conveyed by the total gas curves from mud logging and the pyrograms from pyrolysis-gas chromatography (Py-GC). To improve the accuracy of the predictions, this research proposed a new reservoir-identification methodology that incorporates the graphical information carried by the total gas curves and the pyrograms from Py-GC. The proposed model is applied to Daliuquan Structural Belt to test its performance. Sensitive parameters (the Tg values, the shape of the total gas curve acquired from mud logging, the content of methane in the mud gas, the Pg values, the TPI values, and the patterns of the Py-GC chromatogram) used in this model for reservoir evaluations are determined based on reservoir characterization results. The graphic information carried by the total gas curves from mud logging and the pyrograms from Py-GC is converted into probability problems using multinomial logistic regression in Statistical Package for the Social Sciences (SPSS) software. The well logging data of 137 layers are used to validate the model, and the predictions of the model are compared with the production test data. The overall prediction accuracy of the proposed model is 91.2%. The proposed model is subsequently used to predict the types of 23 reservoirs that are not used in the regression process. The results show that the probability of correctly identifying the types of these reservoirs verified by production tests is 78.3%.