An Improved Probability Particle Swarm Optimization Algorithm

This paper deals with the problem of unconstrained optimization. An improved probability particle swarm optimization algorithm is proposed. Firstly, two normal distributions are used to describe the distributions of particle positions, respectively. One is the normal distribution with the global best position as mean value and the difference between the current fitness and the global best fitness as standard deviation while another is the distribution with the previous best position as mean value and the difference between the current fitness and the previous best fitness as standard deviation. Secondly, a disturbance on the mean values is introduced into the proposed algorithm. Thirdly, the final position of particles is determined by employing a linear weighted method to cope with the sampled information from the two normal distributions. Finally, benchmark functions are used to illustrate the effectiveness of the proposed algorithm.