“…In this study, the following five machine learning algorithms are implemented for dead oil viscosity prediction and results are compared across the board: XGBoost [86], LightGBM [87], random forest [88], an artificial neural network algorithm [89] and SVR [90]. Likewise, due to the superiority and robustness of SuperLearner that is proven in other fields [78,80,[91][92][93], for the first time, this method is also applied, and the output is compared to five other algorithms.…”
Dead oil viscosity is a critical parameter to solve numerous reservoir engineering problems and one of the most unreliable properties to predict with classical black oil correlations. Determination of dead oil viscosity by experiments is expensive and time-consuming, which means developing an accurate and quick prediction model is required. This paper implements six machine learning models: random forest (RF), lightgbm, XGBoost, multilayer perceptron (MLP) neural network, stochastic real-valued (SRV) and SuperLearner to predict dead oil viscosity. More than 2000 pressure–volume–temperature (PVT) data were used for developing and testing these models. A huge range of viscosity data were used, from light intermediate to heavy oil. In this study, we give insight into the performance of different functional forms that have been used in the literature to formulate dead oil viscosity. The results show that the functional form f(γAPI,T), has the best performance, and additional correlating parameters might be unnecessary. Furthermore, SuperLearner outperformed other machine learning (ML) algorithms as well as common correlations that are based on the metric analysis. The SuperLearner model can potentially replace the empirical models for viscosity predictions on a wide range of viscosities (any oil type). Ultimately, the proposed model is capable of simulating the true physical trend of the dead oil viscosity with variations of oil API gravity, temperature and shear rate.
“…In this study, the following five machine learning algorithms are implemented for dead oil viscosity prediction and results are compared across the board: XGBoost [86], LightGBM [87], random forest [88], an artificial neural network algorithm [89] and SVR [90]. Likewise, due to the superiority and robustness of SuperLearner that is proven in other fields [78,80,[91][92][93], for the first time, this method is also applied, and the output is compared to five other algorithms.…”
Dead oil viscosity is a critical parameter to solve numerous reservoir engineering problems and one of the most unreliable properties to predict with classical black oil correlations. Determination of dead oil viscosity by experiments is expensive and time-consuming, which means developing an accurate and quick prediction model is required. This paper implements six machine learning models: random forest (RF), lightgbm, XGBoost, multilayer perceptron (MLP) neural network, stochastic real-valued (SRV) and SuperLearner to predict dead oil viscosity. More than 2000 pressure–volume–temperature (PVT) data were used for developing and testing these models. A huge range of viscosity data were used, from light intermediate to heavy oil. In this study, we give insight into the performance of different functional forms that have been used in the literature to formulate dead oil viscosity. The results show that the functional form f(γAPI,T), has the best performance, and additional correlating parameters might be unnecessary. Furthermore, SuperLearner outperformed other machine learning (ML) algorithms as well as common correlations that are based on the metric analysis. The SuperLearner model can potentially replace the empirical models for viscosity predictions on a wide range of viscosities (any oil type). Ultimately, the proposed model is capable of simulating the true physical trend of the dead oil viscosity with variations of oil API gravity, temperature and shear rate.
“…Jiang [27] et al also proposed a stacked ensemble model, which used RF, ERT, XGBoost, LightGBM, RNN, bidirectional RNN, LSTM, and GRU as basic learners, and used logistic regression as the second-level metalearner. In different application scenarios, the stacked models constructed by researchers such as Li [28], Yang [29], and Feng [30] et al had good results, suggesting an idea for establishing stacked models to predict pavement performance in the field of urban highway tunnel pavement performance prediction.…”
The existing pavement performance prediction methods are limited to single-factor predictions, which often face the challenges of high cost, low efficiency, and poor accuracy. It is difficult to simultaneously solve the temporal, spatial, and exogenous dependencies between pavement performance data and maintenance, the service life of highways, the environment, and other factors. Digital twin technology based on the building information modeling (BIM) model, combined with machine learning, puts forward a new perspective and method for the accurate and timely prediction of pavement performance. In this paper, we propose a highway tunnel pavement performance prediction approach based on a digital twin and multiple time series stacking (MTSS). This paper (1) establishes an MTSS prediction model with heterogeneous stacking of eXtreme gradient boosting (XGBoost), the artificial neural network (ANN), random forest (RF), ridge regression, and support vector regression (SVR) component learners after exploratory data analysis (EDA); (2) proposes a method based on multiple time series feature extraction to accurately predict the pavement performance change trend, using the highway segment as the minimum computing unit and considering multiple factors; (3) uses grid search with the k-fold cross validation method to optimize hyperparameters to ensure the robustness, stability, and generalization ability of the prediction model; and (4) constructs a digital twin for pavement performance prediction to realize the real-time dynamic evolution of prediction. The method proposed in this study is applied in the life cycle management of the Dalian highway-crossing tunnel in Shanghai, China. A dataset covering 2010–2019 is collected for real-time prediction of the pavement performance. The prediction accuracy evaluation shows that the mean absolute error (MAE) is 0.1314, the root mean squared error (RMSE) is 0.0386, the mean absolute percentage error (MAPE) is 5.10%, and the accuracy is 94.90%. Its overall performance is better than a single model. The results verify that the prediction method based on digital twin and MTSS is feasible and effective in the highway tunnel pavement performance prediction.
“…Another new gradient learning framework built up upon the idea of the decision tree is LightGBM 48 . The salient features of LightGBM which dominates XGBoost are consuming less memory, utilizing a leaf-wise growth approach with depth restrictions, and benefiting from a histogram-based algorithm that expedites the training process 49 . Using the aforementioned histogram algorithm, LightGBM discretizes continuous floating-point eigenvalues into k bins, hence leading to building a k-width histogram.…”
Section: Models Implementationmentioning
confidence: 99%
“…The downside of leaf orientation is growing deeper decision trees which unavoidably results in overfitting. However, LightGBM precludes this overfitting while furnishing high efficiency by applying a maximum depth limit to the leaf top 48 , 49 . Figure 2 Leaf-wise tree growth in LightGBM.…”
Due to industrial development, designing and optimal operation of processes in chemical and petroleum processing plants require accurate estimation of the hydrogen solubility in various hydrocarbons. Equations of state (EOSs) are limited in accurately predicting hydrogen solubility, especially at high-pressure or/and high-temperature conditions, which may lead to energy waste and a potential safety hazard in plants. In this paper, five robust machine learning models including extreme gradient boosting (XGBoost), adaptive boosting support vector regression (AdaBoost-SVR), gradient boosting with categorical features support (CatBoost), light gradient boosting machine (LightGBM), and multi-layer perceptron (MLP) optimized by Levenberg–Marquardt (LM) algorithm were implemented for estimating the hydrogen solubility in hydrocarbons. To this end, a databank including 919 experimental data points of hydrogen solubility in 26 various hydrocarbons was gathered from 48 different systems in a broad range of operating temperatures (213–623 K) and pressures (0.1–25.5 MPa). The hydrocarbons are from six different families including alkane, alkene, cycloalkane, aromatic, polycyclic aromatic, and terpene. The carbon number of hydrocarbons is ranging from 4 to 46 corresponding to a molecular weight range of 58.12–647.2 g/mol. Molecular weight, critical pressure, and critical temperature of solvents along with pressure and temperature operating conditions were selected as input parameters to the models. The XGBoost model best fits all the experimental solubility data with a root mean square error (RMSE) of 0.0007 and an average absolute percent relative error (AAPRE) of 1.81%. Also, the proposed models for estimating the solubility of hydrogen in hydrocarbons were compared with five EOSs including Soave–Redlich–Kwong (SRK), Peng–Robinson (PR), Redlich–Kwong (RK), Zudkevitch–Joffe (ZJ), and perturbed-chain statistical associating fluid theory (PC-SAFT). The XGBoost model introduced in this study is a promising model that can be applied as an efficient estimator for hydrogen solubility in various hydrocarbons and is capable of being utilized in the chemical and petroleum industries.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
