Integrating genetic algorithm and support vector machine for polymer flooding production performance prediction

Abstract

Quantitative characterization models of oil increment and water-cut change in polymer flooding called Hou's models are established in the paper. The mathematic models are concise and characteristic parameters have specific physical meanings and are easy to determine. Automatic solution method based on real-coded genetic algorithm (GA) is presented. Based on numerical simulation of polymer flooding, quantitative prediction model of production performance in polymer flooding is established through the combination of orthogonal design and support vector machine (SVM) methods, in which the combination effect of factors is considered. Taking Shengli oilfield as an example, the history matching and prediction of polymer flooding are carried out, it is indicated that there exists a good matching between the quantitative characterization model and the field data, and this model can be extrapolated. Regardless of the limited sample set, the quantitative prediction model can give consideration to both universality and generalization to meet the requirements of engineering computation application. The characterization model or prediction model can be alternatively used according to whether there is a dynamic tendency of the polymer flooding unit or not. Therefore, the models can guide the scheme programming and dynamic adjustments of polymer flooding.