用改进的多分辨小波网络辨识非线性动态系统

Identifying nonlinear dynamic system using the improved multiresolution wave\|net

多分辨小波网络不便于处理非均匀分布样本和在线辨识非线性动态系统 ,对此提出了两点改进 :首先给出一种把非均匀分布样本近似到小波网格系上的方法 ;然后提出一种可变的小波网络及其学习方法 .最后用改进的小波网络辨识非线性动态系统 ,仿真结果验证了这种改进方法的可行性和有效性 .

In most cases, the sampling data in sample space are non-uniformly distributed, which will adversely affect the performance of the multiresolutional wave-nets (MWN). The modified multiresolutional wave-net presented here can efficiently process the non-uniform sampling data. First, the non-uniformly distributed data are located on the grid structure by introducing the rule of approximation location. The error resulted from this locating-grid method is analyzed, and its up-bound is given. Then a variable wave-net is proposed to improve the adaptability of the wave-nets. The structure and parameters of the variable wave-net can be adjusted according to the variation of the practical object. The training speed is remarkably improved, and approximating errors are reduced. Finally, the variable wave-net is used to simulate the identification of a nonlinear dynamic system. The simulation results showed that this improved method is effective and feasible.