Particle Swarm Optimization algorithm easily gets stuck at local optimal solution and shows premature convergence. An improved Particle Swarm Optimization algorithm based on two-subpopulation(TS-IPSO) was proposed. The search range of the algorithm was extended through main subpopulation particle swarm and assistant subpopulation particle swarm, whose search direction was inversed completely. It also adopts the crossbreeding mechanism in genetic algorithm, and uses non-linear inertia weight reduction strategy to accelerate the optimization convergence and improve the search capabilities of particles, then effectively decrease the risk of trapping into local optima. Experiment results have shown that the TS-IPSO can greatly improve the global convergence ability and enhance the rate of convergence, compared with SPSO.