Research output: Contribution to journal › Article › peer-review
Modeling the Vibrational Relaxation Rate Using Machine-Learning Methods. / Bushmakova, M. A.; Kustova, E. V.
In: Vestnik St. Petersburg University: Mathematics, Vol. 55, No. 1, 03.2022, p. 87-95.Research output: Contribution to journal › Article › peer-review
}
TY - JOUR
T1 - Modeling the Vibrational Relaxation Rate Using Machine-Learning Methods
AU - Bushmakova, M. A.
AU - Kustova, E. V.
N1 - Publisher Copyright: © 2022, Pleiades Publishing, Ltd.
PY - 2022/3
Y1 - 2022/3
N2 - Abstract: The aim of the study is to develop an efficient algorithm for solving nonequilibrium gas-dynamics problems in the detailed state-to-state vibrational and chemical kinetics approximation. Optimization of calculation of the vibrational relaxation rate using machine-learning algorithms is discussed. Since traditional methods of calculation require a large number of operations as well as time and memory resources, it is proposed to predict the relaxation rate instead of direct calculations. K-nearest neighbor and histogram gradient boosting algorithms are considered. The algorithms are trained on datasets obtained using two classical reaction-rate coefficient models: the forced harmonic oscillator model and the Schwartz–Slawsky–Herzfeld model. The trained algorithms are used to solve the problem of the spatially homogeneous relaxation of the O2–O mixture. The accuracy and computation time of different methods are compared. It is shown that the used algorithms make it possible to approximate with good accuracy the values of relaxation terms and to solve approximately the system of equations for macroparameters. Based on the obtained data, we recommend the use of machine-learning methods in the problems of nonequilibrium gas dynamics with detailed vibrational and chemical kinetics. Ways of further optimizing the considered methods are discussed.
AB - Abstract: The aim of the study is to develop an efficient algorithm for solving nonequilibrium gas-dynamics problems in the detailed state-to-state vibrational and chemical kinetics approximation. Optimization of calculation of the vibrational relaxation rate using machine-learning algorithms is discussed. Since traditional methods of calculation require a large number of operations as well as time and memory resources, it is proposed to predict the relaxation rate instead of direct calculations. K-nearest neighbor and histogram gradient boosting algorithms are considered. The algorithms are trained on datasets obtained using two classical reaction-rate coefficient models: the forced harmonic oscillator model and the Schwartz–Slawsky–Herzfeld model. The trained algorithms are used to solve the problem of the spatially homogeneous relaxation of the O2–O mixture. The accuracy and computation time of different methods are compared. It is shown that the used algorithms make it possible to approximate with good accuracy the values of relaxation terms and to solve approximately the system of equations for macroparameters. Based on the obtained data, we recommend the use of machine-learning methods in the problems of nonequilibrium gas dynamics with detailed vibrational and chemical kinetics. Ways of further optimizing the considered methods are discussed.
KW - machine learning
KW - nonequilibrium flows
KW - vibrational kinetics
UR - http://www.scopus.com/inward/record.url?scp=85131378924&partnerID=8YFLogxK
UR - https://www.mendeley.com/catalogue/100dae70-945e-36df-a144-28766fb5672c/
U2 - 10.1134/S1063454122010022
DO - 10.1134/S1063454122010022
M3 - Article
AN - SCOPUS:85131378924
VL - 55
SP - 87
EP - 95
JO - Vestnik St. Petersburg University: Mathematics
JF - Vestnik St. Petersburg University: Mathematics
SN - 1063-4541
IS - 1
ER -
ID: 97072550