For reservoir simulation, one of the most important part of reservoir characterization is rock typing, where rock quality is evaluated and estimated for any simulation grid and OOIP (original oil in place) is calculated based on average petrophysical parameters for any layer. To allocate different rock types to simulation grids, rock types should be assigned according to ranges of parameters that differentiate different rock types.

Based on the experience in carbonate reservoirs of XXXX oilfield and other oilfields, irreducible water saturation (Swi) is a critical differentiation parameter for rock typing, although it can be difficult and expensive to evaluate. In oil zones, water saturation from log data is assumed to be the irreducible water saturation. The problem arises in transition zone and water zone, where water saturation from log data is not equal to the irreducible water saturation of that rock.

KNN(K-Nearest Neighbor) is an effective machine learning method for classification and regression in many industries including geo-science. Models can be trained and predict irreducible water saturation from the traditional logs such as GR, Density, Neutron, Sonic using KNN and other Machine Learning methods using labelled data from oil zones. Randomly selected 50% of the dataset was used for training and other 50% was used as testing dataset to be predicted. The prediction precision of KNN method can reach the minimum 92% line for all 25 wells studied and is most robust compared to other methods such as Random Forest and SVM. The trained model was used to predict all the rock types in the reservoir and was confirmed in wells with core data and other advanced measurements data.

A new approach of petrophysical rock typing (PRT) for carbonate reservoir using KNN based on traditional wireline data and core analysis data is studied and the results show it can solve the PRT problems in carbonate reservoir simulation without acquiring extra data and additional cost. A new workflow was established to process wireline data and provide the PRT results based on wireline data for every newly drilled well on top of traditional "Porosity-Permeability-Saturation" petrophysical evaluation results. This paper presents the methodology, workflow, results, verification, as well as appropriate application scenarios of this new approach. Considering the requirements of the data input and the workflow of the approach, it could be applied widely in similar carbonate reservoirs.

You can access this article if you purchase or spend a download.