Abstract: To predict the system nonlinear and non-stationary conditions more accurately, a method of chaos prediction based on chaotic neural networks is introduced to predict the velocity of the air-defense rocket launcher AC servo system, which paves the way for the trend prediction of system nonlinear and non-stationary conditions. C-C method is used to select the proper embedding dimension and time delay. The phase space of system is reconstituted using experimental data of system’s irregular movement and is analyzed. The context neurons of output layer are added to the original Elman network, and the self-feedback gain coefficients are trained as connective weight, which could strengthen the nonlinear approximation ability of Elman network. Then a model of chaotic neural network based on the improved Elman network is set up. The predictions based on the maximun Lyapunov exponent and chaotic neural network are performed, respectively. The predicted results show that the prediction based on the chaotic neural network has a higher accuracy, which makes the trend forecasting of the system more effectively.

Key words: ordnance science and technology, rocket launcher, servo system, state forecasting, phase-space reconstruction, chaotic neural network