Data-Driven approaches to optimize chemical kinetic models

The data-driven optimized mechanism has recently been proposed for chemical kinetic modeling of the combustion of real, multi-component fuels. Experimental datasets of ignition delay times across wide temperature and equivalence ratio ranges are obtained by using a rapid compression machine (RCM) and shock tube. The HyChem (Hybrid Chemistry) and lumped NTC (Negative Temperature Coefficient) approaches have been used to model chemical reactions under high and low temperature chemistry, respectively. The reaction coefficients including pre-exponential factors, corrective coefficients, and activation energies are optimized against empirical results. This paper employs and compares three different types of heuristic optimization techniques: a micro-genetic algorithm, a Bayesian optimization, and a stochastic gradient descent (SGD). We demonstrate the approaches in HyChem-oriented chemical kinetic models for multi-component fuels, Jet A. The results show that all techniques are capable of optimizing the chemical kinetics models, but computational costs and performance vary among the different approaches.