This paper first discusses the structure, principle and learning algorithm of Elman neural network model. A modified Ehnan neural network model is then proposed by adding new adjustable weights between the context nodes and the output nodes to enhance its dynamical character. The corresponding learning algorithm is also derived by using steepest descent principle. Theoretical analysis and simulation results show that this kind of modified Ehnan neural network learns much faster than the original model.
Hunt K J, Sbarbaro D, Zbikowski R, et al.Neural networks for control system-A survey[J].Automatica.1992, 28(6):1083-1112[2]Narendra K S, Parthasarathy K. Identification and control of dynamical systems using neural networks[J].IEEE Trans. on Neural Networks.1990, 1(1):4-27[3]高晓智,王常虹,徐立新,等.CMAC神经网络再励学习控制.CIAC95中国智能自动化学术会议里智能自动化专业委员会成立大会论文集,天津:1995,638-643.[4]Elman J. Finding structure in time[J].Cognitive Science.1990, 14(2):179-211[5]Sastry P S, Santharam G, Unnikrishnan K P. Memory neuron networks for identification and control of dynamical systems[J].IEEE Trans. on Neural Networks.1994, 5(2):306-319