Boost-phase guidance with neural network for interception of ballistic missile

In order to meet the special control requests of an intercept missile, a new boost-phase guidance law is proposed based on the theory of pesudospcetral method and artificial neural network. A group of optimal trajectories with multiple constraints, obtained by hp-adaptive pesudospcetral method, is used as samples to train the neural network. To show the effect of training patterns on the guidance performance, three training patterns with different input and output vectors are studied in this paper. The new guidance law which the neural network is used to generate attitude command turns out to be the best solution for the problem here, compared to the traditional training pattern. It eliminates the drawback effect of flight-path angle on the guidance performance, so that sufficient robustness is obtained. Moreover, it has a smaller miss distance while achieving larger final velocity. The simulation results show that the performance of the new guidance law is very close to the optimal trajectory, and more suitable for the real-time application considering the ability of sensors.