This thesis deals with numerical analysis and practical implementation of pursuit-evasion games. As the need for developing a guidance law that provides good interception performance against an intelligent target, increases, pursuit-evasion game becomes an attractive concept in missile guidance. For the sake of implementing game solutions into real situations, it is indispensable to build a reliable numerical method able to take into account realistic engagement situations, as well as to construct a feedback guidance law from the numerical solutions. This thesis proposes a co-evolutionary method for solving pursuit-evasion games and a neural network feedback guidance law based on game solutions.
The gradient-based method provides very precise numerical solutions for pursuit-evasion games. It contains some shortcomings, though, of difficulty in initial guess, and incapability of considering lethal radius or miss distance. In this thesis, a co-evolutionary method is proposed to supplement these shortcomings of the gradient-based method with numerical verification. In addition, the neural network is employed in order to build a feedback. guidance law from the numerical game solutions obtained by using the gradient-based method. Accompanying with performance enhancing techniques for the neural network guidance, numerical simulations are performed to compare the proposed guidance law with proportional navigation.