论文部分内容阅读
在基于神经网络的系统辨识中,传统辨识方法易陷于局部寻优,或在优化过程中忽略了网络权值和结构的关联性而进行单一优化,这都难以建立精确模型。针对此问题,本文采用遗传算法浮点数编码的方式,同时优化神经网络权值及拓扑结构,设计评估综合性能的适应度函数,求出最优解。最后通过试验仿真,证明此方法的可行性。
In the system identification based on neural network, the traditional identification method is easy to be trapped in the local optimization, or single optimization is ignored in the optimization process by ignoring the correlation between network weights and structure, which makes it difficult to establish an accurate model. In order to solve this problem, this paper adopts genetic algorithm floating-point coding and optimizes the weights and topology of neural network. The fitness function to evaluate the overall performance is designed and the optimal solution is obtained. At last, the feasibility of this method is proved by experimental simulation.