Appl. Sci. 2024, 14(19), 9035; https://doi.org/10.3390/app14199035 (registering DOI) - 7 Oct 2024
Abstract
Accurately predicting the filtration volume (FV) in drilling fluid (DF) is crucial for avoiding drilling problems such as a stuck pipe and minimizing DF impacts on formations during drilling. Traditional FV measurement relies on human-centric experimental evaluation, which is time-consuming. Recently, machine learning
[...] Read more.
Accurately predicting the filtration volume (FV) in drilling fluid (DF) is crucial for avoiding drilling problems such as a stuck pipe and minimizing DF impacts on formations during drilling. Traditional FV measurement relies on human-centric experimental evaluation, which is time-consuming. Recently, machine learning (ML) proved itself as a promising approach for FV prediction. However, existing ML methods require time-consuming input variables, hindering the semi-real-time monitoring of the FV. Therefore, employing radial basis function neural network (RBFNN) and multilayer extreme learning machine (MELM) algorithms integrated with the growth optimizer (GO), predictive hybrid ML (HML) models are developed to reliably predict the FV using only two easy-to-measure input variables: drilling fluid density (FD) and Marsh funnel viscosity (MFV). A 1260-record dataset from seventeen wells drilled in two oil and gas fields (Iran) was used to evaluate the models. Results showed the superior performance of the RBFNN-GO model, achieving a root-mean-square error (RMSE) of 0.6396 mL. Overfitting index (OFI), score, dependency, and Shapley additive explanations (SHAP) analysis confirmed the superior FV prediction performance of the RBFNN-GO model. In addition, the low RMSE (0.3227 mL) of the RBFNN-NGO model on unseen data from a different well within the studied fields confirmed the strong generalizability of this rapid and novel FV prediction method.
Full article
(This article belongs to the Topic Petroleum and Gas Engineering)
►
Show Figures
Figure 1
Figure 1
<p>Flowchart of the procedure applied for developing simple and hybrid ML models to predict FV in DF from two easily measured input variables, FD and MFV.</p> Full article ">Figure 2
<p>Applying the GO algorithm to determine optimal hyperparameter values for the RBFNN predictor.</p> Full article ">Figure 3
<p>Flowchart illustrating the implementation of the MELM-GO hybrid ML model designed for predicting the FV properties of DFs.</p> Full article ">Figure 4
<p>Heatmap correlation matrix displaying the relationships between independent and dependent parameters for the compiled dataset.</p> Full article ">Figure 5
<p>RMSE results for the FV dataset evaluated with the RBFNN model with three training/testing subset separation ratios.</p> Full article ">Figure 6
<p>Identification of outliers in the FV training dataset using GPR–Mahalanobis distance (MD) modeling.</p> Full article ">Figure 7
<p>The iterative reduction of RMSE within various iterations of the GO algorithm is employed to ascertain the optimal structure of the MELM algorithm in the prediction of the FV.</p> Full article ">Figure 8
<p>Comparative cross-plots evaluating accuracy of measured and predicted FV values utilizing (<b>a</b>) MELM, (<b>b</b>) MELM-GO, (<b>c</b>) RBFNN, and (<b>d</b>) RBFNN-GO models on training data.</p> Full article ">Figure 8 Cont.
<p>Comparative cross-plots evaluating accuracy of measured and predicted FV values utilizing (<b>a</b>) MELM, (<b>b</b>) MELM-GO, (<b>c</b>) RBFNN, and (<b>d</b>) RBFNN-GO models on training data.</p> Full article ">Figure 9
<p>Assessment of the error convergence of GO algorithm iteration sequences for the two HML models configured to predict the FV.</p> Full article ">Figure 10
<p>Comparative cross-plot accuracy evaluations of measured and predicted FV values achieved by the trained (<b>a</b>) MELM, (<b>b</b>) MELM-GO, (<b>c</b>) RBFNN, and (<b>d</b>) RBFNN-GO models applied to the testing data subset.</p> Full article ">Figure 11
<p>Radar chart contrasting prediction scores achieved by standalone ML and HML models.</p> Full article ">Figure 12
<p>Relationships between the percentage of added noise to the input variable distributions and R<sup>2</sup> values for FV predictions for the ML and HML models.</p> Full article ">Figure 13
<p>Visualizing the effect of the two input features on the FV predictions with SHAP values for the RBFNN-GO model applied to the training subset: (<b>a</b>) SHAP detailed feature impact plot and (<b>b</b>) SHAP summary plot of the feature importance.</p> Full article ">Figure 14
<p>The (<b>a</b>) 3D and (<b>b</b>) 2D heat map partial dependence plots showcasing the interplay between pairs of input features in the predictions of the FV as generated by the RBFNN-GO model applied to the training subset.</p> Full article ">Figure 15
<p>Comparison of the measured FV values with those predicted by the RBFNN-GO model for the unseen data.</p> Full article ">Figure 16
<p>Workflow diagram demonstrating how the configured HML models can be applied in the DF well-site laboratory to assist the drilling crew with FV semi-real-time monitoring and decision making.</p> Full article ">
<p>Flowchart of the procedure applied for developing simple and hybrid ML models to predict FV in DF from two easily measured input variables, FD and MFV.</p> Full article ">Figure 2
<p>Applying the GO algorithm to determine optimal hyperparameter values for the RBFNN predictor.</p> Full article ">Figure 3
<p>Flowchart illustrating the implementation of the MELM-GO hybrid ML model designed for predicting the FV properties of DFs.</p> Full article ">Figure 4
<p>Heatmap correlation matrix displaying the relationships between independent and dependent parameters for the compiled dataset.</p> Full article ">Figure 5
<p>RMSE results for the FV dataset evaluated with the RBFNN model with three training/testing subset separation ratios.</p> Full article ">Figure 6
<p>Identification of outliers in the FV training dataset using GPR–Mahalanobis distance (MD) modeling.</p> Full article ">Figure 7
<p>The iterative reduction of RMSE within various iterations of the GO algorithm is employed to ascertain the optimal structure of the MELM algorithm in the prediction of the FV.</p> Full article ">Figure 8
<p>Comparative cross-plots evaluating accuracy of measured and predicted FV values utilizing (<b>a</b>) MELM, (<b>b</b>) MELM-GO, (<b>c</b>) RBFNN, and (<b>d</b>) RBFNN-GO models on training data.</p> Full article ">Figure 8 Cont.
<p>Comparative cross-plots evaluating accuracy of measured and predicted FV values utilizing (<b>a</b>) MELM, (<b>b</b>) MELM-GO, (<b>c</b>) RBFNN, and (<b>d</b>) RBFNN-GO models on training data.</p> Full article ">Figure 9
<p>Assessment of the error convergence of GO algorithm iteration sequences for the two HML models configured to predict the FV.</p> Full article ">Figure 10
<p>Comparative cross-plot accuracy evaluations of measured and predicted FV values achieved by the trained (<b>a</b>) MELM, (<b>b</b>) MELM-GO, (<b>c</b>) RBFNN, and (<b>d</b>) RBFNN-GO models applied to the testing data subset.</p> Full article ">Figure 11
<p>Radar chart contrasting prediction scores achieved by standalone ML and HML models.</p> Full article ">Figure 12
<p>Relationships between the percentage of added noise to the input variable distributions and R<sup>2</sup> values for FV predictions for the ML and HML models.</p> Full article ">Figure 13
<p>Visualizing the effect of the two input features on the FV predictions with SHAP values for the RBFNN-GO model applied to the training subset: (<b>a</b>) SHAP detailed feature impact plot and (<b>b</b>) SHAP summary plot of the feature importance.</p> Full article ">Figure 14
<p>The (<b>a</b>) 3D and (<b>b</b>) 2D heat map partial dependence plots showcasing the interplay between pairs of input features in the predictions of the FV as generated by the RBFNN-GO model applied to the training subset.</p> Full article ">Figure 15
<p>Comparison of the measured FV values with those predicted by the RBFNN-GO model for the unseen data.</p> Full article ">Figure 16
<p>Workflow diagram demonstrating how the configured HML models can be applied in the DF well-site laboratory to assist the drilling crew with FV semi-real-time monitoring and decision making.</p> Full article ">
