Normalized Depths as Key Input and Detailed QC Steps for Improved Permeability Predictions Using Existing Machine Learning Techniques
- Deepak Kumar Voleti (Abu Dhabi Company for Onshore Petroleum operations Ltd.) | Ashish Kundu (Abu Dhabi Company for Onshore Petroleum operations Ltd.) | Maniesh Singh (Abu Dhabi Company for Onshore Petroleum operations Ltd.)
- Document ID
- Society of Petroleum Engineers
- Abu Dhabi International Petroleum Exhibition & Conference, 13-16 November , Abu Dhabi, UAE
- Publication Date
- Document Type
- Conference Paper
- 2017. Society of Petroleum Engineers
- 5.6 Formation Evaluation & Management, 7.6.6 Artificial Intelligence, 5.8.7 Carbonate Reservoir, 5 Reservoir Desciption & Dynamics, 7.6 Information Management and Systems, 1.6 Drilling Operations, 5.6.4 Drillstem/Well Testing, 5.8 Unconventional and Complex Reservoirs, 6 Health, Safety, Security, Environment and Social Responsibility, 1.6.9 Coring, Fishing, 6.1.5 Human Resources, Competence and Training, 7.6.4 Data Mining, 5.5.2 Core Analysis, 6.1 HSSE & Social Responsibility Management, 7 Management and Information
- Ktest/Kmodel, Normalized Depth, Permeability prediction, Machine learning, QC plots
- 3 in the last 30 days
- 115 since 2007
- Show more detail
- View rights & permissions
|SPE Member Price:||USD 8.50|
|SPE Non-Member Price:||USD 25.00|
Because of advance computing technologies, machine learning today is widely used for many types of prediction. The performance of existing prediction techniques has been quite acceptable but still it needs improvement. In oil industry, this type of machine learning is being used for Permeability prediction. In case of homogeneous reservoir, conventional permeability prediction techniques are very well documented in the available published literature and are easy to be implemented however, it is a real challenge in case of complex carbonate reservoir which are highly affected by diagenetic overprint. The machine learning algorithms exactly does according to what user had trained for it. It get confused either if it receives a new unseen input for which it was not trained or it has multiple answer to the same input. This study shows that by using selective intelligent inputs (quantitative factors influencing permeability) and with proper segregated and intuitive training to machine it is possible to gain better correlation between different inputs and a reliable permeability prediction can be achieved.
The studied Early Cretaceous reservoir has a large areal extent with 20% average porosity and large range of permeability, from 0.1 to 1000 md due to preferential diagenetic processes over large variety of lithofacies. Capturing high permeability streaks/zone simultaneously with low K intervals was quiet important to understand the fluid flow dynamics and hence was a real challenge in this project. This paper describe the workflow developed to analyze data in different dimensions away from the conventional ways, detailed QC steps and preparing the intelligent input data sets (Normalized depths, Rock types, zonal and sectorial flags, synthetic permeability guide curve, etc.) which facilitate the machine to make a best decision and output optimum results. This paper shows two- point normalization method to bring reservoir depth interval between 0 to100 and pick up consistent permeability profile in each sector which correlates better to stratigraphic framework.
In this study, the models were built with 3081 CCA data from 30 spatially distributed cored wells and blind tested on 10 other cored wells; later these models were used to predict permeability in 210 uncored wells. Comparison of the predicted permeability with the well test permeability shows Ktest/Klog in the range of 1.05 to1.6 suggesting fair to very good reconciliation. The normalized depth vs. predicted permeability and an independent cumulative permeability QC plots are introduced to qualify the prediction model.
The workflow shown in this paper is logically built and gives the interpreter flexibility to choose inputs based on the observed permeability variance. The QC flags and the predictive models are easy to implement and update permeability model as the new data comes.
|File Size||3 MB||Number of Pages||19|
Ghedan S.G,Weldu T.,Al-Farisi O. (2010): Hybrid Log-Derived Permeability Prediction Model for a Heterogeneous Carbonate Reservoir with Tarmat Layers Considering Different Levels of Cutoffs. SPE-138685-MS. DOI: https://doi.org/10.2118/138685-MS
Shokir E. M. El-M. (2004): A Novel Model for Permeability Prediction in Uncored Wells. SPE-87038-MS. DOI: https://doi.org/10.2118/87038-MS
Barman I.,Sharma A.K,Walker R.F.,Datta A.G. (1998): Permeability Predictions in Carbonate Reservoirs Using Optimal Non-parametric. SPE-39667-MS. DOI: https://doi.org/10.2118/39667-MS
Stundner M.,Oberwinkler C. (2004): Self-Organizing Maps for Lithofacies Identification and Permeability Prediction. SPE-90720-MS. DOI: https://doi.org/10.2118/90720-MS
Rebelle M., AL Neaimi M.,Ribeiro M.T.,Gottlib-Zeh S.,Valsardieu B.,Moss B. (2005): Quantitative and Statistical Approach for a New Rock- and Log-Typing Model: Example of Onshore Abu Dhabi Upper Thamama. IPTC-10273-MS. DOI: https://doi.org/10.2523/IPTC-10273-MS
Rebelle M. (2014): Rock-typing In Carbonates: A Critical Review Of Clustering Methods. SPE-171759-MS. DOI: https://doi.org/10.2118/171759-MS
Terrilyn M.,Olson (1998): Porosity and Permeability Prediction in Low-Permeability Gas Reservoirs from Well Logs Using Neural Networks. SPE-39964-MS. DOI:https://doi.org/10.2118/39964-MS
Warren J.E,Skiba F.F.,Price H.S. (1961): An Evaluation of the Significance of Permeability Measurements. SPE-1641-G-PA. DOI: https://doi.org/10.2118/1641-G-PA