Evaluation of petrophysical classification of strongly heterogeneous reservoirs based on the MRGC algorithm

The target formation in the study area of the Pearl River Mouth Basin is characterized by complex lithology and thin interbedded layers, with a large pore-permeability distribution range and strongly heterogeneous characteristics, which makes the reservoir pore structure and production capacity significantly different and brings research difficulties for reservoir logging evaluation and desert identification. The conventional reservoir classification method is mainly based on physical research, which requires developing extremely accurate formulas for calculating porosity and permeability; the calculation accuracy of pore permeability of low-porosity and low-permeability reservoirs is difficult to guarantee; and the conventional logging data cannot be comprehensively applied in reservoir classification. In this paper, taking Zhujiang and Zhuhai Formation reservoirs in the Huizhou M oilfield as an example, we integrated core analysis data such as core cast thin section, pore permeability data, rock electrical parameters, grain size, and relative permeability curves and combined with petrophysical parameters and pore structure characteristics to classify the reservoirs. The artificial neural network is used to predict the resistivity of saturated pure water (R-0) to remove the influence of oil and gas on reservoir resistivity. The natural gamma ray (GR) "fluctuation" is used to calculate the variance root of variation (G(S)) to reflect the lithological variability and sedimentary heterogeneity of the reservoir, and then the conventional logging preferences, R-0 and Gs (based on GR), are classified based on the automatic clustering MRGC algorithm to classify the logging facies. To classify the petrophysical phase reservoirs under the constraint of pore structure classification, we proposed a petrophysical classification logging model based on the natural gamma curve "fluctuation" intensity for strongly heterogeneous reservoirs. The learning model is extended to the whole area for training and prediction of desert identification, and the prediction results of the model are in good agreement with the actual results, which is important for determining favorable reservoirs in the area and the adjustment of oilfield development measures.