AN EFFICIENT PROXY-MODEL TO PREDICT THE PERFORMANCE OF ENHANCED OIL RECOVERY – POLYMER FLOODING METHOD

• Main Author: Prayoga, Advarel
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/40066
• Institution: Institut Teknologi Bandung

This paper presents a novel approach to generate proxy-model that is sufficiently good in predicting cumulative oil production (Np) of polymer flooding without conducting numerical simulation. The model can be used as preliminary evaluation to determine optimum scenario and the suitable polymer criteria that can give maximum profit in polymer flooding project. The study was focused on the practical aspects of polymer flooding that are necessary to be understood to generate efficient proxy-model and how the result model can be improved. Proxy-model was generated by conducting several simulations with various parameters and the result of simulation will be modeled by several regression method. To make efficient proxy-model, understanding the physical phenomena and collecting historical project record of polymer flooding are necessary to determine the parameterization range value, therefore the proxy-model can be used practically. In this study, CMOST was used to automatically run 5842 cases to capture numerical model’s behavior. As this study play with big and robust data, sufficient knowledge of data science is necessary to have predictive model with sufficient R-Square. CMOST was then used to build proxy-model using polynomial regression method and it shows quite good value of R-Square, that is 0.9445. Even though the R-Square showed a quite good value, plot of real value - predict value showed dissatisfaction. Hence, model was then enhanced using another statistical modelling tool called Eureqa. Eureqa, an AI-powered modeling engine, was used to formulate predicted value with real value using genetic algorithm. Enhanced proxy can be approached by two ways: formulate all data in one go or formulate all data by clustering it first. The result from both approaches were compared with CMOST proxy-model using Mean Absolute Percentage Error (MAPE) analysis. It showed that cluster formulization was better with MAPE of 10.71% by only 5.2 barrel/day as its highest absolute error. In this study, results of the model would be reassessed until the quality is eligible to be used. By improving the model quality, this study presents a great potential for proxy-model to predict Np. Moreover, the suggested methodology can be used for further proxy-model development. The critical thing in providing efficient proxy-models is it can be easily used and has the ability to reduce computational load in doing simulation. Hence, this method can improve the work efficiency, cost, and time.