A Novel Adaptive Non-Linear Regression Method to Predict Shale Oil Well Performance Based on Well Completions and Fracturing Data

Abstract This paper presents the results of applying a novel nonlinear regression method, Variable Structure Regression (VSR), to forecasting well performance given the well completion and rock composition data. We compiled and analyzed data from 79 producing wells from the same unconventional reservoir. Calibration using the performance data of 60 wells was used to predict the behavior for the rest, and the predictions are quite successful. Input parameters for the regression model were the number of frac stages, the average length of each stage, isochore, total organic content, proppant-to-fluid ratio and a rock brittleness metric derived from illite content. The cumulative oil production after three months and after 18 months was considered as outputs of the prediction model. Scatterplot analysis did not indicate any obvious correlations between the individual input parameters and the output, thereby necessitating the use of a more complex multi-parameter model. Given the relatively small training data size and the complexity of the problem, the VSR method achieved satisfactory prediction accuracy. For predicting 3-month cum oil, model calibration using the performance data of 60 wells was used to predict the behavior for the remaining 19. About 70% of the predictions were within a 30% margin of error. For predicting 18-month cum oil, data from 33 wells was used to predict the production of 10 wells. About 80% of the predictions for the 18-month cum production were within the 30% error margin.