A modified elman neural network according to the dynamic & kinematic equations of helicopters was established, the training algorithms for the neural networks(NN) was derived, and the training steps and their realizations were provided. From the telemetry data collection of real trial flights of a type of helicopter, the data of typical flying attitudes were exploited to train the NN and calculate the elements of associated weighted matrix, thus the NN for dynamic analysis against the helicopter was obtained. To illustrate the NN effeteness, the vertical operation response in the constant velocity forward flying attitude was simulated, comparing simulation results with real trial flight data shows that the NN basically reflects the real dynamic behavior of the helicopter.
�ܹ���.����ʽֱ�������ж���ѧ��ģ����ֵģ�� .����:�������պ����ѧ���տ�ѧ�빤��ѧԺ,2003 Zhou Guoyi. Modeling and numerical simulation of thedynamics of coaxial helicopter . Beijing:School of Aeronautic Science and Engineering, Beijing University of Aeronautice and Astronautics, 2003(in Chines)
Hornik K, Stinchornbe M, White H. Multilayer feedforward network are universal approximators[J].Neural Network.1989,2:359-366
��ƽ��,�ų�ˮ.�˹���������ģ���������[M].����:�廪��ѧ������,2005:276-327 Yan Pingfan, Zhang Changshui. Artificial neural networks and evolutionary computing[M]. Beijig:Tshinghua University Press, 2005:276-327(in Chinese)
Williams R, Zipser D.A learning algorithm for continually running fully recurrent NN[J].Neural Computation.1989,1:270-280
Gernasky M. Simple recurrent network trained by RTRL and extended Kalmann filter algorithm[J]. Neural Network World, 2003, 13(3):223-234