Nonlinear System Identification Using DAFNN
A new artificial neuron model was presented according to the mechanism of biological neurons. Because of a dynamic activation function, it was regarded as a dynamic nonlinear mapping. A new network called DAFNN, the abbreviation of Dynamic Activation Function Neural Network, of such neurons was also proposed. It could be used as dynamic plant identifiers or controllers. The performance of DAFNN was verified by comparing it with PIDNN, a neural network is composed of P, I and D neurons, and NARXNN, also called Nonlinear AutoReguressive eXgenous input filter.Simulation results show the performance of DAFNN was better than that of PIDNN. Without adding computation complexion, DAFNN could replace NARX recurrent neural network in nonlinear plant modeling and identification.
Neuron Model Activation Function Nonlinear System Identification
Ming Li Chengwu Yang
College of Power Engineering, Nanjing University of Science & Technology Nanjing, 210094, China
国际会议
杭州
英文
961-965
2006-10-12（万方平台首次上网日期，不代表论文的发表时间）