Machine Learning-Based Models for Basic Sediment & Water and Sand-Cut Prediction in Matured Niger Delta Fields

Abstract

Oil production from matured fields in the Niger Delta is characterised by basic sediment and water (BS&W) and sand or sand-cut (Scut) production. The predominant factor for this production is the unconsolidated nature of the formations in the Niger Delta. The available correlations for estimating BS&W and Scut are based more on the intrinsic reservoir properties than controllable wellhead variables during oil production. This study developed neural-based models to predict BS&W and Scut based on multiple-inputs single-output (MISO) and multiple-inputs multiple-outputs (MIMO) networks using 457 datasets from 43 oilfields in the Niger Delta. The performances of the neural-based models with new fields test datasets were determined using some statistical yardsticks: coefficient of determination (R2), correlation coefficient (R), mean square error (MSE), root mean square error (RMSE), average relative error (ARE), and average absolute relative error (AARE). The results indicate that the MISO neural-based models had overall R and MSE values of 0.9999 and 2.0698\(\times\)10-5, respectively, for BS&W and 0.9995 and 2.1529\(\times\)10-6 for Scut. In contrast, the MIMO neural-based model had overall R and MSE values of 0.9997 and 7.5865\(\times\)10-5. The generalisation performance of the MISO neural-based models with new field test datasets resulted in R2, R, MSE, RMSE, ARE and AAPRE of 0.97406, 0.98695, 2.08143, 1.44272, -0.00638 and 0.28755, respectively, for the BS&W model and R2 of 0.89558, R of 0.93544, MSE of 0.01736, RMSE of 0.13177, ARE of 0.01338 and AARE of 0.01759 for the Scut model. Furthermore, the MIMO-based model with new field test datasets resulted in R2, R, MSE, RMSE, ARE and AAPRE of 0.97317, 0.98650, 2.15293, 1.46729, -0.00713 and 0.25064, respectively, for BS&W, while the Scut model had R2 of 0.87505, R of 0.93544, MSE of 0.02118, RMSE of 0.14554, ARE of -0.02280 and AARE of 0.02996. Also, the relative importance of the input parameters of the MISO and MIMO neural-based models in predicting BS&W and Scut is \(q_0\) >Pr>Pwh>S> \(\gamma\)API . Based on the statistical indicators obtained, the predictions of the developed neural models were close to the actual fields’ datasets. Thus, the neural-based models should apply as tools for estimating BS&W and Scut in mature fields in the Niger Delta.