Application of an improved particle swarm optimization algorithm for neural network training

Particle Swarm Optimization (PSO) is an evolutionary computation technique developed by Kennedy and Eberhart in 1995 and has been applied successfully to various optimization problems. The PSO idea is inspired by natural concepts such as fish schooling, bird flocking and human social relations. It combines local search (by self experience) and global search (by neighboring experience), possessing high search efficiency. Backpropagation (BP) is generally used for neural network training. It is very important to choose a proper algorithm for training a neural network. In this paper, we present a modified particle swarm optimization based training algorithm for neural network. The proposed method modify the trajectories (positions and velocities) of the particle based on the best positions visited earlier by themselves and other particles, and also incorporates population diversity method to avoid premature convergence. Experimental results have demonstrated that the modified PSO is a useful tool for training neural network.