Evaluation of a genetic algorithm that modifies air traffic data for conflict probe testing

Air traffic scenarios based on recorded live data are essential for the development, testing and evaluation of air traffic automation. It is often desirable to modify the live data to introduce additional encounters and conflicts because live data generally contains no conflicts. To this end a genetic algorithm (GA) has been developed that time shifts individual flights in the scenario based on a set of characteristics. Although this approach has been successful, larger scenarios and additional constraints dramatically increase the time for the GA to reach an acceptable solution. This paper first introduces an enhanced GA technique that distributes the processing over several computers. Following the biological metaphor further, this approach is called the Island Model because each instance of the running algorithm represents an island where periodic sharing of information represents migration between islands. The paper concludes with an evaluation of the Island Model utilizing a set of designed experiments. The evaluation will not only consider time to solve but the quality of the solution produced relative to the specified constraints, with the goal of identifying important factors and optimal settings for those factors.