Hybrid machine learning algorithms to predict condensate viscosity in the near wellbore regions of gas condensate reservoirs

Dades bibliogràfiques
Parent link:Journal of Natural Gas Science and Engineering
Vol. 95.— 2021.— [104210, 26 p.]
Autor corporatiu: Национальный исследовательский Томский политехнический университет Инженерная школа природных ресурсов Отделение нефтегазового дела
Altres autors: Behesht Abad Abouzar Rajabi, Mousavi Seyedmohammadvahid M., Mohamadian N. Nima, Wood D. А. David, Ghorbani Hamzeh, Davoodi Sh. Shadfar, Ahmadi A. M. Alvar Mehdi, Shahbazi Kh. Khalil
Sumari:Title screen
Gas condensate reservoirs display unique phase behavior and are highly sensitive to reservoir pressure changes. This makes it difficult to determine their PVT characteristics, including their condensate viscosity, which is a key variable in determining their flow behavior. In this study, a novel machine learning approach is developed to predict condensate viscosity in the near wellbore regions ([mu]c) from five input variables: pressure (P), temperature (T), initial gas to condensate ratio (RS), gas specific gravity ([gamma]g), and condensate gravity (API). Due to the absence of accurate recombination methods for determining [mu]c machine learning methods offer a useful alternative approach. Nine machine learning and hybrid machine learning algorithms are evaluated including novel mul-tiple extreme learning machine (MELM), least squares support vector machine (LSSVM) and multi-layer per-ceptron (MLP) and each hybridized with a particle swarm optimizer (PSO) and genetic algorithm (GA). The new MELM algorithm has some advantages including 1) rapid execution, 2) high accuracy, 3) simple training, 4) avoidance of overfitting, 5) non-linear conversion during training, 6) no trapping at local optima, 6) fewer optimization steps than SVM and LSSVM. Combining MELM with PSO, to find the best control parameters, further improves its performance. Analysis indicates that the MELM-PSO model provides the highest μc predic-tion accuracy achieving a root mean squared error (RMSE) of 0.0035 cP and a coefficient of determination (R2) of 0.9931 for a dataset of 2269 data records compiled from gas-condensate fields around the world. The MELM-PSO algorithm generates no outlying data predictions. Spearman correlation coefficient analysis identifies that P, [gamma]g and Rs are the most influential variables in terms of condensate viscosity based on the large dataset studied.
Режим доступа: по договору с организацией-держателем ресурса
Idioma:anglès
Publicat: 2021
Matèries:
Accés en línia:https://doi.org/10.1016/j.jngse.2021.104210
Format: Electrònic Capítol de llibre
KOHA link:https://koha.lib.tpu.ru/cgi-bin/koha/opac-detail.pl?biblionumber=665379

MARC

LEADER 00000naa0a2200000 4500
001 665379
005 20250910102649.0
035 |a (RuTPU)RU\TPU\network\36578 
035 |a RU\TPU\network\33974 
090 |a 665379 
100 |a 20210921d2021 k||y0rusy50 ba 
101 0 |a eng 
102 |a NL 
135 |a drcn ---uucaa 
181 0 |a i  
182 0 |a b 
200 1 |a Hybrid machine learning algorithms to predict condensate viscosity in the near wellbore regions of gas condensate reservoirs  |f Behesht Abad Abouzar Rajabi, M. Mousavi Seyedmohammadvahid, N. Mohamadian [et al.] 
203 |a Text  |c electronic 
300 |a Title screen 
330 |a Gas condensate reservoirs display unique phase behavior and are highly sensitive to reservoir pressure changes. This makes it difficult to determine their PVT characteristics, including their condensate viscosity, which is a key variable in determining their flow behavior. In this study, a novel machine learning approach is developed to predict condensate viscosity in the near wellbore regions ([mu]c) from five input variables: pressure (P), temperature (T), initial gas to condensate ratio (RS), gas specific gravity ([gamma]g), and condensate gravity (API). Due to the absence of accurate recombination methods for determining [mu]c machine learning methods offer a useful alternative approach. Nine machine learning and hybrid machine learning algorithms are evaluated including novel mul-tiple extreme learning machine (MELM), least squares support vector machine (LSSVM) and multi-layer per-ceptron (MLP) and each hybridized with a particle swarm optimizer (PSO) and genetic algorithm (GA). The new MELM algorithm has some advantages including 1) rapid execution, 2) high accuracy, 3) simple training, 4) avoidance of overfitting, 5) non-linear conversion during training, 6) no trapping at local optima, 6) fewer optimization steps than SVM and LSSVM. Combining MELM with PSO, to find the best control parameters, further improves its performance. Analysis indicates that the MELM-PSO model provides the highest μc predic-tion accuracy achieving a root mean squared error (RMSE) of 0.0035 cP and a coefficient of determination (R2) of 0.9931 for a dataset of 2269 data records compiled from gas-condensate fields around the world. The MELM-PSO algorithm generates no outlying data predictions. Spearman correlation coefficient analysis identifies that P, [gamma]g and Rs are the most influential variables in terms of condensate viscosity based on the large dataset studied. 
333 |a Режим доступа: по договору с организацией-держателем ресурса 
461 |t Journal of Natural Gas Science and Engineering 
463 |t Vol. 95  |v [104210, 26 p.]  |d 2021 
610 1 |a электронный ресурс 
610 1 |a труды учёных ТПУ 
610 1 |a hybrid machine learning algorithms 
610 1 |a gas condensate viscosity 
610 1 |a multi-layer extreme learning machine 
610 1 |a multilayer perceptron 
610 1 |a least squares support vector machine 
610 1 |a алгоритмы 
610 1 |a машинное обучение 
610 1 |a вязкость 
610 1 |a газовые конденсаты 
701 0 |a Behesht Abad Abouzar Rajabi 
701 1 |a Mousavi Seyedmohammadvahid  |b M. 
701 1 |a Mohamadian  |b N.  |g Nima 
701 1 |a Wood  |b D. А.  |g David 
701 1 |a Ghorbani  |g Hamzeh 
701 1 |a Davoodi  |b Sh.  |c specialist in the field of petroleum engineering  |c Research Engineer of Tomsk Polytechnic University  |f 1990-  |g Shadfar  |3 (RuTPU)RU\TPU\pers\46542  |9 22200 
701 1 |a Ahmadi  |b A. M.  |g Alvar Mehdi 
701 1 |a Shahbazi  |b Kh.  |g Khalil 
712 0 2 |a Национальный исследовательский Томский политехнический университет  |b Инженерная школа природных ресурсов  |b Отделение нефтегазового дела  |3 (RuTPU)RU\TPU\col\23546 
801 2 |a RU  |b 63413507  |c 20210921  |g RCR 
856 4 |u https://doi.org/10.1016/j.jngse.2021.104210 
942 |c CF