Evolutionary Algorithms for the Large-scale Business Tourist Problem

Several novel modifications of speciated genetic algorithms have been proposed: random predator prey assignment, maximizing entropy while preserving balance between predator and prey assignment (further referred to as optimal assignment), optimal assignment with the species' fitness being the solution of the SLAE, optimal assignment with recalculating the individuals' fitness after solving SLAE for species' fitness. New modifications have been tested on the four known instances of TSP. Several mutation and crossover operators have been used while testing. New modification have successfully beaten the classical GA with the same number of fitness evaluations on all benchmark but rat195 with DPX crossover. Modified GA have also been compared with classical GA with a bigger number of fitness evaluations and was able to robustly beat the classical GA on eil51 instance. We have argued that the modified GA has a higher probability to converge to a better optimum than the classical GA that can be easily stuck in a local optimum. Modified GA is noticed to have lower variance in the final result in comparison with classical GA, so fewer launches are needed to be convinced that the current solution is close enough to the best solution that can be obtained with the given GA.