Abstract:  Particle swarm optimization(PSO)algorithm is a populatiorrbased search strategy, which has exhibited good per-
formance in dealing with difficult optimization tasks. However,as a stochastic optimization algorithm,PSO tends to suffer from
premature convergence on most problems,which is resulted from the swarm losing its diversity.To solve the problem of being
trapped in local minima owing to the loss of the diversity in the optimi}ztion process,an improved particle swarm optimi}ztion
based on gradient search is proposed in this paper.The proposed method improves its search performance from two aspects. On
one hand ,in the attraction phase,centering around the global best particle,particles search linearly flowing in their negative gra-
dicnt directions. When the particle is very close to the neighborhood of the global minima,its velocity is decreased little by little
by a similar dichotomy method ,and the global search is well performed. On the other hand ,in the repulsion course,the repulsion
velocities of the particles are adjusted,according to the diversity of the swarm. The more the diversity is,thc less the repulsion
velocities decrcase,and vice-versa,and thus the population activity is maintained.The proposed algorithm could not only retain
swarm diversity,but also possess strong local search ability as well as global search ability. Experimental results verify that the
proposed algorithm has better convergence performance and more robust property than basic PSO and some corresponding im
provcmcnts,cspccially on finding minima of multimodal and high dimension functions.