Adaptive Simulated Binary Crossover with Bayesian Optimization for Industrial Applications

Genetic Algorithms are widely used as global optimization solvers for complex and non-convex problems in industrial applications (resource and load planning, workshop logistics, process control, etc.). Simulated Binary Crossover (SBX) is a common operator in real-coded Genetic Algorithms for continuous optimization, but it typically relies on a fixed distribution index parameter that may not stay optimal throughout the evolutionary process. To address this, we propose a Transfer Learning-Based Adaptive Simulated Binary Crossover approach, which employs Bayesian Optimization to dynamically update the parameter during the Genetic Algorithm run. The parameter is periodically re-optimized based on the population's current dynamics. Tested on benchmark functions including Rosenbrock, Rastrigin, Ackley, and Griewank, our method achieves faster convergence and higher solution quality than standard Genetic Algorithms with a fixed parameter.