An Application of Proxy-Modeling Framework for Numerical PVT-Models Matching on Laboratory Measured Data

摘要

The paper presents a new automated method of matching the equation of state (EOS) of multicomponent system to the laboratory measured data. The method is based on the idea of using a statistical estimator, which allows you to replace the entire space of possible PVT-models that correspond to different parameter values by a simplified, piecewise-polynomial proxy-model. During the application of the method, the proxy model is refined based on the results of intermediate calculations and a global search for the optimal solution is perfomed. The new approach avoids the drawbacks of the traditional approach using the linear regression or gradient descent methods. The reason is that in highly non-linear problem traditional methods often finds only a local minimum and does not allow us to obtain a solution with an acceptable degree of accuracy. The proposed method allows to effectively adjust the PVT-model to actual data, and mostly times, with higher accuracy than the traditional method. The application of this method to adjust the equation of state for laboratory measured data is shown by the example of one of the gas condensate field sample. This allowed to reduce the PVT-model error more than 5 times in comparison with the result of using linear regression.