Model-structure analysis of dynamic neural networks with particle-swarm optimization
微粒群优化动态神经网络模型结构分析

Due to the random uncertainty in the particle-swarm optimization algorithm(PSO), the analysis of stability is difficult to be performed. Most of the researchers on PSO solve this problem based on the practical model obtained by experience. Being different, we employ the robust uncertainty theory to decompose the original algorithm into the timeinvariant part and the uncertain time-variant part for reducing the original fixed constraints on parameters, and perform the asymptotic stability analysis by using the PSO algorithm with dynamic inertia weight. By using the Lyapunov method, we obtain the sufficient conditions of stability for the dynamic neural networks based on PSO and the upper and lower bounds of the parameters to be adjusted, providing the theoretical basis for parameter selection. Finally, simulation examples validate the stability conditions and the effectiveness of the proposed dynamic neural networks based on PSO algorithm.