A genetic algorithm based anti-submarine warfare simulator

• Main Author: Timmerman, Michael Jay.
• Other Authors: Shing, Man-Tak
• Language: en_US
• Published: Monterey, California. Naval Postgraduate School 2013
• Online Access: http://hdl.handle.net/10945/26571

This research was aimed at improving the genetic algorithm used in an earlier anti-submarine warfare simulator. The problem with the earlier work was that it focused on the development of the environmental model, and did not optimize the genetic algorithm which drives the submarine. The improvements to the algorithm centered on finding the optimal combination of mutation rate, inversion rate, crossover rate, number of generations per turn, population size, and grading criteria. The earlier simulator, which was written in FORTRAN-77, was recoded in Ada. The genetic algorithm was tested by the execution of several thousand runs of the simulation, varying the parameters to determine the optimal solution. Once the best combination was found, it was further tested by having officers with anti-submarine warfare experience run the simulation in various scenarios to test its performance. The optimum parameters were found to be: population size of eight, five generations per turn, mutation rate of 0.001, inversion rate of 0.25, crossover rate of 0.65, grading criteria of sum of the fitness values of all alleles while building the strings, and checking the performance against the last five environments for the final string selection. The use of these parameters provided for the best overall performance of the submarine in a variety of tactical situations. The submarine was able to close the target and execute an attack in 73.1% of the two hundred tests of the final configuration of the genetic algorithm