.png){kind=icon}
.png){kind=icon}
Return
Transient search driven random forest model for predicting diluted heavy crude oil viscosity
.png){kind=icon}
PRE
AI
A
J
DOI：10.3389/fphy.2026.1775533.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Viscosity is a basic thermophysical property that determines the flow characteristics of single phase oil such as crude oil. Accurate estimation of diluted heavy crude oil viscosity is important for laboratory characterization and operational handling of heavy oils. In this research, a ML based approach is presented for estimating the viscosity of diluted heavy crude oil samples measured in laboratory conditions. A total of 245 experimental datasets were collected using a Brookfield DV2T viscometer. In this experiment, heavy crude oil was mixed with lighter oil at different dilution rates and temperatures. The major input parameters considered are heavy crude oil viscosity, lighter oil viscosity, dilution rate and temperature. To improve the performance of the model, Min-Max normalization was used for data scaling and Kernel Principal Component Analysis (KPCA) was used for nonlinear feature extraction. A Transient Search driven Random Forest Regression (TS-RFR) model was introduced to optimize hyperparameters and enhance predictive performance. The performance of the proposed model was assessed using Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Squared Error (MSE), coefficient of determination (R 2) and Percentage of Accuracy-Precision (PAP). The comparative analysis with the existing ML models, such as MLP-ANN, SVR and LightGBM, reveals that the proposed TS-RFR model performs better with RMSE = 0.2976, MAE = 0.1005, MSE = 0.1015, R 2 = 97.35% and PAP = 92.35%. The analysis clearly shows that the proposed model is a reliable and efficient tool for estimation of diluted crude oil viscosity.
Keywords:
crude oil viscosity
diluted heavy oil
estimation
kernel principal component analysis (KPCA)
min max normalization
transient search driven random forest regression (TS-RFR)
Journal
F
IF:
2.1
Papers:
200
Citations:
0
Organization
Cited Papers