Combining Recursive Least Square and Principal Component Analysis for Assisted History Matching

History matching is a process of altering parameters in a reservoir simulator in order to match production performance with observed historical data. There are two methods of history matching; manual history matching, which is the most common approach, and another method is assisted history matching. However, most of the reservoirs in reality are heterogeneous and it requires experience and time to rely on trial and error methods. Assisted history matching can be an alternative solution in saving time thus, using optimization process needs to be further developed and improved to be implemented.
The main objective of this project is to investigate the applicability of Recursive Least Squares (RLS) for parameter estimation methods in assisted history matching. Currently not much attention is given for using RLS for history matching purposes. Even though RLS is a simple and effective method to estimate parameters, RLS have stability problem when number of parameters is high. Therefore, in this project, Principal Component Analysis (PCA) is used to reduce the number of parameters.
The project is divided to several steps; in which the first step is to develop a conceptual model which can be used to generate both synthetic historical data and also simulated data. Forward model was also involved in the process of defining the objective function. Next, using simulated data together with historical data, objective function will be computed. This project will study the applicability of the combined algorithm for history matching problem.
The study conducted on PCA and RLS method shows high chances of success in applying these methods for history matching problem. The algorithm formulated also can easily be practiced, provided with ample knowledge of numerical computational tool to implement it. When RLS and PCA are applied, the result obtained with estimated parameter results in lower mean squared error (MSE) between historical and matched result which is 0.75% compared to MSE between historical and simulated which is 5.17%. This proves that when RLS is applied, almost 5% of error can be reduced and thus can result in better forecasting of the reservoir production performance.