Quantitative characterization and application of deep learning carbonate reservoir

Abstract

Abstract Quantitative reservoir characterization is an important research task in carbonate reservoir development. At present, the comprehensive evaluation of drilling, logging, seismic, and productivity testing data is usually used, which requires a large number of test data and tedious manual data collation. Deep learning technology can independently learn complex reservoir characteristics, has a strong ability for high-dimensional data structure mining, and has a strong advantage in solving complex nonlinear reservoir description problems closely related to various geological characteristics. Therefore, this paper applies deep learning technology to reservoir characterization. First, a deep neural network model is constructed, and then seven characteristic parameters such as fracture density, fracture opening, fracture porosity, matrix porosity, total porosity, effective thickness, and specific oil production index are taken as input variables. After model training and verification, the effective permeability distribution map of the reservoir in the target area can be automatically generated. The results show that the deep learning method has strong adaptability and practicability for the quantitative characterization of complex reservoirs, and provides a new way of thinking for the quantitative characterization of reservoirs.