| The development of carbonate reservoirs is closely related to sedimentary facies. The identification of
carbonate rock fabric components with logging data is of great practical significance for the study of carbonate
sedimentary facies. However, due to the strong diagenesis and the development of fractures and vugs in the carbonate
strata, it has always been a difficult problem to use conventional logging data to accurately identify lithofacies based on
rock fabric classification. At present, the combination of effective logging parameters and machine learning methods has
become an effective means to improve the identification accuracy. Among them, Lucia’s apparent rock fabric number
(ARFN) parameter has achieved good application results in carbonate rocks. However, ARFN technology is established in
non-fractured and non-water zones, which limits its application. Therefore, this paper takes the Cambrian Longwangmiao
Formation in the GM area of Sichuan Basin as an example, and proposes an improved method. The study area is
dominated by carbonate ramp deposition, which is controlled by late dolomitization and hypergene dissolution, with
dissolution pores and multiphase fractures developing. The reservoir has good physical properties but strong
heterogeneity. Firstly, with reference to the Lucia’s rock type classification scheme, the lithofacies are divided into three
categories: granular dolomite (including sandy dolomite and finely crystalline dolomite), very finely crystalline dolomite
and micrite dolomite (including micrite dolomite and argillaceous micrite dolomite). Then, through the relationship
analysis of the rock types and log response, the most sensitive curves are selected as density, acoustic wave and natural
gamma curve. Based on the logging porosity, logging water saturation and thin section identification data, Lucia’s ARFN
formula is improved according to the conditions of water layer and non-water layer. The results show that ARFN curve
can quickly and quantitatively identify rock fabric components, and the recognition coincidence rate is high for granular
dolomite and micrite dolomite, but low for very finely crystalline dolomite. Finally, the improved ARFN is used as one of
the logging input parameters, and combined with the K-neighbor classification algorithm (KNN) to further improve the
accuracy of lithofacies logging identification. The core data verification shows that the average coincidence rate increases
from 74% to more than 80%, which effectively promotes the fine study of carbonate sedimentary microfacies of the
Longwangmiao Formation in the GM area of Sichuan Basin. The improved ARFN formula can be applied to any nonmicrobial
carbonate strata, especially for dolomite strata with fewer rock types, to achieve rapid and quantitative
identification of lithofacies, so this technology is suitable for other similar carbonate rocks. The lithofacies logging
identification of the formation has reference and promotion value. However, ARFN technology is established in nonfractured
and non-aqueous layers, which limits its application. |
