Maximizing the Utilization of Well Log Data to Deliver Realistic Permeability Model by Combining Neural Network and Geostatistical Approach
(Case Study: BBT Field, Tunisia)
Sugiyanto S., Nouha B., Charfeddine M., Lotfi Fourati, Zul Anwar, and M. Fakhrur Razi.
Permeability model is one of the most important parameter on reservoir modeling since it will affect the simulation result in the perspective of reservoir productivity and field development plan. The most common method used to predict the permeability value is by performing transform, correlating core permeability data with one or two affecting parameters. This transform commonly will result on trends of permeability of each defined facies as a function of porosity and/or Vshale. And at the end of modeling process, the trend will be directly used to convert the porosity to permeability.
That common method has one major drawbacks of using only limited parameters for prediction. In complex reservoir rock deposits, permeability value might not be affected by porosity and Vshale only, but also by the mineralogy of the rock, by grain density, or even by the fluid saturation. In that case, it is impossible to find the perfect trend of permeability that can respect the scattered trend of core data. Accordingly, the common permeability transform tend to oversimplify the real condition of the reservoir.
To solve that common problem, ideally the permeability value should be predicted based on multi-parameters transform. However, manual approach of multi-parameter transform is considered to be impractical due to the difficulties of finding specific mathematical equation of each reservoir. For that reason, automatic brute-force approach i.e. neural network is suggested to perform the trending. Finally, the 3D distribution can still be implemented using common geostatistical approach.
This method requires two parameters to be introduced as intermediate result i.e. permeability trend (k-trend) and residual permeability (Dlogk). The k-trend is initially calculated by using the global trend and then corrected by residual permeability that is predicted from well log. Generalized procedures are as follows: (i) performs global trending (porosity-permeability relationship); (ii) calculates k-trend based on the global trend formula; (iii) calculates residual permeability Dlogk=log(k)-log(k-trend); (iv) predicts residual permeability for un-cored well & un-cored interval based on well log data using neural network; (v) calculates 3D model k-trend from distributed porosity model; (vi) distributes Dlogk for 3D model using geostatistical method; and (vii) calculates the final 3D permeability k=10^(log(k-trend)+ Dlogk).
The procedure has successfully been implemented in BBT Field study to predict the reservoir permeability distribution. Permeability trend is initially calculated based on porosity, and finally corrected by the residual permeability that is estimated based on log parameters i.e. GR, RHOZ, THPN, PEFZ, PHIE, VCL, and zone. The final permeability model shows very good representation of core data in view of trend as well as the range of the actual core permeability.
Key words: permeability, porosity, transform, artificial neural network
Download full paper in here