Evolutionary generation of artificial neural network based guidance laws

In this paper AI techniques are applied to the production of air-to-air missile guidance laws. A Genetic Algorithm (GA) is used as a numerical optimisation tool to optimise neural network based guidance laws. Traditionally techniques including Optimal Control Theory, Feedback Linearisation and Differential Game Theory have been applied to the problem. Using Genetic Algorithms one can incorporate constraints, a more realistic engagement model and noise models with more ease hut, at the expense of computational requirements. Using a modified Genetic Algorithm and a planar air-to-air missile simulation a neural network based guidance law has been evolved. This law has been evolved to work in a large range of engagement geometries and against a. range of target, manoeuvres with robustness issues in mind. Their performance has been assessed by comparison to a PN and a DGT based guidance law. Simulations results indicate the artificial neural network guidance law is more effective than these two guidance laws. The robustness of the neural network guidance laws has also been demonstrated against synthetic targets with Radar Cross Sections (RCS).