Statistical Racing Crossover Based Genetic Algorithm for Vehicle Routing Problem

Genetic algorithms are modular metaheuristics simulating the evolutionary process over a solution set. The optimization is very adaptive but slow, making statistical research difficult. In this paper an algorithm is proposed where different variants are racing against each other while statistics are gathered. Our results show that this algorithm is an efficient, standalone, and even more adaptive solution. Those variants that result in faster convergence lead the race, but get stuck in local minima. In these cases, the more agile combinations with slower convergence gain higher probability and find better solutions farther from the local minimum. The hybrid is capable of faster convergence with minimal additional runtime. We also provide complexity estimations for resource requirements.