Oil Well Flow Rates and Components Fraction Prediction Using Nonparametric Variable Similarity Based Modeling

Abstract The Variable Similarity Based Modeling and Virtual Signal Generation algorithms are applied to continuously predict (infer) the values of flow rates and component fractions in between periodic oil well tests. This method uses the time-varying signals collected during irregular well tests to train a nonparametric regression model which represents the current operating state of the well. Additionally, this approach also provides the benefit of early warning on potential well performance degradation or other anomalies. The proposed approach was shown of being able to predict the oil, gas and water flow with relatively small errors (~3%, ~3% and ~10%, respectively).