Machine Learning Algorithms for Parameter Identification for Reactive Flow in Porous Media

Earlier we have explored different deterministic, stochastic and metaheuristic methods for identifying parameters of heterogeneous reactions for diffusion dominated and reaction dominated regimes [2,3,4,5]. Pore scale reactive transport was studied, breakthrough curves were the additional information used in identifying the parameters in Henry or Langmuir isotherms. All methods were time consuming, requiring multiple solution of the direct problem. Various surrogate models are used in the literature to reduce the computational burden related to parameter identification problems. In this paper we explore surrogate models based on neural network, Gaussian process, and cross approximation approaches. We also extend the number of the sought parameters. The achieved accuracy and the performance of the surrogate models were studied.