Abstract: Particle swarm optimization (PSO) is a novel swarm intelligence algorithm inspired by certain social behavior of bird flocking originally. Since proposed in 1995, the algorithm proved to be a valid optimization technique and has been applied in many areas successfully. However, like others evolutionary algorithms, PSO also suffers from the premature convergence problem, especially for the large scale and complex problems. In order to alleviate the premature convergence problem, the paper develops a self-organized PSO(SOPSO). SOPSO regards the swarm as a self-organized system, and introduces negative feedback mechanism to imitate the information interaction between the particles and the swarm background. Considering swarm diversity is a key factor influencing the global performances of PSO, SOPSO adopts swarm diversity as main dynamic information to control the tuning of parameters through feedback, which in turn can modify the particles to diverge or converge adaptively and contribute to a successful global search. The proposed methods are applied to some complex function optimizations and compared with the other notable improved PSO. Simulation results show SOPSO based on feedback control of swarm diversity is a feasible technique, which can alleviate the premature convergence validly and improve the global performances of PSO in solving the complex functions.