Optimizing Chemical Reaction Mechanisms: Evaluating Parameter-Free Metaheuristic Algorithms and Gradient-Based Optimization

Abstract

ABSTRACT In the present work, we compare the performance of four different metaheuristic algorithms in reaction mechanism optimization problems. Mechanism optimization is an essential step to develop detailed and skeletal reaction mechanisms. Stochastic metaheuristic algorithms that mimic biological evolution, such as the genetic algorithm (GA), are often used for this purpose. While this technique has proven effective over a broad class of problems, it requires the specification of many hyperparameters or user-defined inputs, such as the crossover rate, mutation rate, offspring selection criterion, among others. In the present work, we show that extensive knowledge about the underlying system is needed to ensure efficient performance from the GA. Particle swarm optimization (PSO) is another metaheuristic technique that has gained popularity in the chemical kinetics community. However, it also suffers from sensitive dependence on the input parameters. The performance of two parameter-free algorithms, the Teaching-Learning-Based Optimization (TLBO) and the Artificial Bee Colony (ABC) algorithm are compared against GA and PSO. The TLBO and ABC algorithms have not been tested previously in reaction mechanism optimization problems to the best knowledge of the authors. We show that TLBO provides comparable or superior performance than other optimization strategies in terms of fitness value, computational time, recovery of the original coefficients, and the prediction of the species mole fractions. Additionally, as the method is hyperparameter-free, the sensitive dependence on user-defined input is eliminated. In the last section of the study, the performance of a gradient-based optimization algorithm (Adam) is compared against the metaheuristic algorithms. The performance of Adam is sensitive to the learning rate, which is a hyperparameter for the algorithm. The effectiveness of TLBO method for the optimization of reaction mechanisms is established.