基于免疫遗传算法的递归模糊神经网络

为了解决递归网络的梯度信息不易获取而传统遗传算法训练时间过长、易于早熟的问题,提出了一种用于辨识非线性动态系统的递归高木-关野模糊神经网络(T_RFNN:Takagi-Sugeno Recurrent Fuzzy Neural Network).T_RFNN是在高木-关野模糊模型的基础上加入了反馈层,利用免疫遗传算法对T_RFNN的参数进行训练和调整.该网络具有更少的网络参数、更快的收敛速度和更高的精度等特点,能够很好地完成动态非线性系统的映射.与高木-关野模糊神经网络相比,网络参数减少了45%,网络误差减少了65%,而网络的运行时间提高了近68%.T_RFNN仿真实验的辨识结果也表明,该网络在训练次数明显减少的情况下学习性能得到了显著改善.

Recurrent Fuzzy Neural Networks Based on Immune Genetic Algorithm

In order to solve the problem of the grads information acquirement and the prematurity of tradition genetic algorithm with long train time, a T--RFNN(Takagi-Sugeno Recurrent Fuzzy Neural Network ) for identifying nonlinear dynamic systems is proposed. The T--RFNN combines the recurrent multilayered connectionist network with dynamic TS(Takagi-Sugeno) fuzzy model. Based on an IMG(Immune Genetic Algorithm), the training parameters in the T--RFNN are adjusted. the mapping of nonlinear dynamic systems is commendably finished. The proposed T--RFNN has a smaller network structure, a smaller number of network parameters, and a faster convergence speed. Compared with the TS fuzzy neural networks, network parameters are reduced to 55% of the former, network error is reduced to 35% of the former and the run time is increased by 68% approximately in the T--RFNN. The simulative experimental results show that it has obvious improvement in learning performance with the case of fewer train time.