PSO and RBF network-based Wiener model and its application to system identification

In this paper, a new kind of Wiener model structure is introduced, which is realized by using the mapping function of neural networks. The model uses the linear dynamic neurons and a RBF network to express one Wiener model´s dynamic linear part and static nonlinear part respectively. The parameter identification for the new Wiener model adopts the unified identification method. The learning of parameters includes two cycles that the inner-cycle is executed by gradient training methods based on the BP thought and the outer-cycle uses the PSO (Particle Swarm Optimization) algorithm. The training method based on unified identification makes the new Wiener model converge to the steady state along the expected direction with a small error in a short time. The Wiener model is applied to the identification of the famous Box and Jenkins gas CO₂ density, and the simulation results show that the method proposed in this paper is effective.