An improved continuous particle swarm optimization (PSO) algorithm was proposed to solve the overlapping coalition generation problem. During the process of solution, an integer based encoding strategy was proposed which could represent the resource distribution of each agent to every task coalition, and the validation and revision algorithms for particle encoding were also presented. In order to improve the diversity of the swarm, a twostage initialization method of particle position was proposed. In the evolution process, the swarm trapping at the stagnation was reinitialized around its weighted centroid position, which could guide the swarm to avoid local extremum and accelerate the convergence of the algorithm. Contrastive experiments indicate that the proposed algorithm is effective.