Developing a Computational Framework to Harness Genetic Algorithms for Enhanced Optimization of Bioprocesses

Resumen

The optimization of bioprocesses frequently involves highly nonlinear dynamic models that challenge derivative-based methods. In this study, a kinetic model of xylose fermentation for xylitol production was optimized with a hierarchical genetic-algorithm (GA) framework. First, a secondary GA minimized an economic objective function that penalized residual substrate and rewarded product formation; the Runge Kutta Fehlberg (RKF45) scheme solved the model under a constant local-error tolerance. Subsequently, a primary GA tuned the secondary GA’s population size and crossover fraction. The secondary GA alone revealed a “stability valley” at crossover fractions: 0.60 ≤ cf ≤ 0.90, where convergence proved steady and standard deviations remained low. Benchmarks with the two-dimensional Ackley function confirmed that solution surface ruggedness, rather than dimensionality, governed GA performance. The hierarchical approach identified an optimal configuration: cf = 0.53 and population = 260, which lay outside the stability valley and elevated the objective function by 3.1 % relative to the best value within the valley (0.6 ≤ cf ≤ 0.9). No modifications were introduced to the kinetic model or to the economic criterion; the gain therefore arose exclusively from metaparameter tuning. However, computation time increased, indicating that hybrid strategies, such as variable-fidelity models or adaptive-mutation schemes, may be required for larger systems. Even so, the demonstrated improvement translates into meaningful economic benefits and underscores the value of systematic GA meta-optimization for industrial bioprocesses. The results provide a reproducible benchmark and a foundation for extending this framework to more complex models, hybrid metaheuristics, and machine-learning-guided parameter adjustment.