Factorial analysis of a real time optimisation for pursuit-evasion problem

An optimal three dimensional pursuit-evasion (PE) problem between two aerial vehicles is applied using parallel evolutionary programming (PEP) algorithm. The evader is modeled as a six degrees of freedom generic jet fighter aircraft with aileron, elevator, rudder and throttle setting as the control variables. The pursuer is a medium range generic air-to-air missile and modeled as a point-mass body. The game is played in three dimensions. The pursuer seeks to intercept the evader and the evader seeks to avoid interception. The search for an optimal evading solution is performed by using parallel evolutionary programming (PEP). The optimisation algorithm has to minimise the objective function, which is a ¡¯selfplay¡¯ pursuit-evasion simulation. The value of the game is the outcome of the ¡¯self-play¡¯ game, i.e. capture or no capture. The solution that gives the maximum value of fitness with no capture is considered the best. The good solution is found to be highly dependent on the initial condition, the number of generation, the size of population and the number of CPUs used. The complex interaction between parameters is best studied from a statistical perspective. The interaction between parameters is studied by using factorial analysis. Three main parameters are studied: the size of the population, the number of generation and the number of CPUs used. The outcome is the number of good solutions found and the time taken to find the first optimal solution.