An improved Quantum-behaved particle swarm optimization algorithm for training fuzzy neural networks

Abstract

Quantum-behaved particle swarm optimization (QPSO), which is inspired by analysis of the convergence of the traditional PSO and quantum system, is a global optimization algorithm. In this paper, we propose an improved QPSO, namely iQPSO, with a generating process of initial population of particles and a mutation operator for training fuzzy neural networks (FNNs). Traditionally, the parameters of FNNs are trained by gradient-based methods, and it may fall into a local minimum. In iQPSO, the generating process of initial population can speed up the convergence. The mutation operator diversifies the population and prevents premature convergence to local minima. The robotic path planning is adopted to demonstrate the proposed method. Simulation results have shown that the performance of proposed iQPSO is superior to other types of QPSO.