APPEA Journal 2010 50th ANNIVERSARY ISSUE—567 A NOVEL APPROACH IN EXTRACTING PREDICTIVE INFORMATION FROM WATER-OIL RATIO FOR ENHANCED WATER PRODUCTION MECHANISM DIAGNOSIS Lead author Minou Rabiei M. Rabiei 1 , R. Gupta 1 ,Y.P. Cheong 2 and G.A. Sanchez Soto 2 1 Department of Mathematics and Statistics Curtin University of Technology Kent St, Bentley Perth WA 6102 2 CSIRO, Earth Science and Resource Engineering 26 Dick Perry Ave Kensington WA 6151 Minou.rabiei@postgrad.curtin.edu.au R.gupta@curtin.edu.au Yawpeng.cheong@csiro.au Gerardo.sanchezsoto@csiro.au ABSTRACT Despite the advances in water shutoff technologies, the lack of an efficient diagnostic technique to identify excess water production mechanisms in oil wells is preventing these technologies being applied to deliver the desired results, which costs oil companies a lot of time and money. This paper presents a novel integrated approach for diagnosing water production mechanisms by extracting hidden predictive information from water-oil ratio (WOR) graphs and integrating it with static reservoir parameters. Two common types of excess water production mechanism (coning and channelling) were simulated where a wide range of cases were generated by varying a number of reservoir parameters. Plots of WOR against oil recovery factor were used to extract the key features of the WOR data. Tree-based ensemble classifiers were then applied to integrate these features with the reservoir parameters and build classification models for predicting the water production mechanism. Our results show high rates of prediction accuracy for the range of WOR variables and reservoir parameters ex- plored, which demonstrate the efficiency of the proposed ensemble classifiers. Proactive water control procedures based on proper diagnosis obtained by the proposed tech- nique would greatly optimise oil productivity and reduce the environmental impacts of the unwanted water. KEYWORDS WOR plots, excess water production, channelling, con- ing, classification, random forests, ensemble techniques. INTRODUCTION In recent years, unconventional mathematical tech- niques and soft computing methodologies have gained more and more popularity in the oil and gas industry. The complex nature of the oil fields combined with staggering volume and diversity of data and resulting uncertainties calls for more sophisticated techniques to integrate vari- ous types of data, quantify uncertainties, identify hidden patterns and extract useful information. Data mining is one of the promising methodologies that can offer great benefits to the oil industry by extracting hidden predictive information from the large and/or complex databases.This technique uses past and present information to discover previously unknown patterns in the data, and then train and build models to predict future trends and behavior (Kantardzic, 2002). Classification trees are one of the most popular classification algorithms used in data mining. Classification trees are powerful knowledge models that predict the value of a target variable based on several in- put variables. They are easy to use, simple to understand and interpret, and require little data preparation (Tomei, 2008). Nevertheless, they do not always provide the most accurate result. A simple and effective procedure to tackle this deficiency is to use an ensemble of classifiers instead of using a single, large and less accurate tree classifier (Kuncheva, 2004). Classifier ensembles are aggregations of several classifiers (either different types of classifiers or different instants of the same classifier), whose indi- vidual predictions are combined in some manner (e.g., averaging or voting) to form a final prediction (Oza and Tumer, 2008). Because they use all the available classifier information, ensembles generally provide better and more robust solutions in most applications. In this paper we investigate the application of such en- semble techniques to the classification and prediction of excess water production mechanisms in vertical oil wells. Excess water production is a serious economic and envi- ronmental problem in most mature oil fields. Accurate and timely diagnosis of the water production mechanism is critical in the success of the applied well treatment meth- odology. Incorrect, inadequate, or lack of proper diagnosis usually leads to ineffective water control treatments that cost a lot of time and money (Seright et al, 2001). Many em- pirical techniques such as decline curve plots, and water-oil ratio (WOR) versus cumulative oil production or time have traditionally been used in production data analysis (Poe