An improved particle swarm optimization using best neighbor with worst particle and its application in soft-sensor of gasoline endpoint

Abstract

This paper proposes out a variation of particle swarm optimization with best neighbor and worst particle (BNWPPSO). In BNWPPSO, some particles will be constructed as new neighbors of each particle and the best one of them will have influence on the behavior of the particle. The update formula of position is modified also to balance the local search ability and global search ability more efficiency. The worst particle of the swarm will be re-randomized at every generation to prevent premature convergence of PSO. BNWPPSO is investigated by several benchmark problems, the results show that BNWPPSO performances better than traditional PSO. Furthermore, BNWPPSO is applied to train artificial neural network to construct a soft-sensor of gasoline endpoint of crude distillation unit. The results show that the model constructed by BNWPPSO is feasible and effective.