Integrating More Hard Data for Sedimentary Facies Modeling: A New Algorithm Based on Neural Network and Krigging Theories

Facies are important in reservoir modeling because the petrophysical properties of interest are highly correlated with facies type. Knowledge of facies constrains the range of variability in porosity, permeability and... [ view full abstract ]

Facies are important in reservoir modeling because the petrophysical properties of interest are highly correlated with facies type. Knowledge of facies constrains the range of variability in porosity, permeability and especially saturation functions. Sequential indicator simulation (SIS) is widely used for facies modeling because the results have high variability and may be well characterized with anisotropy. However, very limited hard data is applied for this method—commonly only facies type data identified from well logging or core combined with geological trend and seismic data is used. A new method based on neural network and krigging algorithms (NKA) is proposed in this paper to integrate more hard data for facies modeling and analyze the uncertainty of the result. Necessary steps for NKA include: (1) Prepare related data, which includes GR, SP, AC logs (or more logs), RMS seismic amplitude data and facies codes (representing different facies types) for all wells; (2) Construct quantitative function between facies codes and these logs and RMS amplitude with BP neural network algorithm, using half of these well data; (3) Verify this function with the remaining half sets of well data and obtain the function's accuracy rate; (4) Construct the spatial distribution of GR, SP and AC values and calculate their correlated variances for each cell by krigging method (SK or OK) and extract RMS seismic amplitude value at each node; (5) Calculate facies code for each cell with the function obtained in (2); (6) Assuming Gaussian distribution for GR, SP and AC values at each cell, the preliminary probability of the facies code being present at the current location is mimicked using that function by Monte Carlo simulation, and the final probability can be obtained when the accuracy rate of that function is considered. Finally, facies type and its probability are determined at each cell, with more hard data involved. The prevailing advantage of NKA over SIS is concerning more hard data for calculation. The relationship between lithology parameters measured in logs and facies types on wellbores is expanded for areal facies modeling. Seismic trend and facies models can be honored and NKA can also be used for simulation. NKA and SIS facies modeling techniques are both applied in a block of Dagang oilfield in the east of China, which is dominated by braided river sediments. The results of NKA are more continuous and realistic and have much less outliers compared to indicator krigging.