Volterra system-based neural network modeling by particle swarm optimization approach

In this paper, we propose a novel identification method for nonlinear discrete dynamic systems. A feedforward neural network with the structure of Volterra system is newly presented. This kind of mathematical model possesses more adjustable parameters than the original Volterra system to further enhance the modeling capacity. Tuning adjustable parameters inside the neural network is based on the particle swarm optimization (PSO) instead of the commonly used back-propagation method. The PSO algorithm is with multiple direction searches and can easily find out the global solution for the given optimization problem. This paper also develops the whole design steps for PSO-based feedforward neural network modeling for nonlinear discrete systems. Two kinds of examples are illustrated to validate the feasibility and efficiency of the proposed method. In addition, some examinations containing different initial conditions and population sizes, and presence of measurement noises are considered to evaluate the modeling performance.