An adaptive Particle Swarm Optimization based on cloud model theory is proposed to improve its capability and applied to minimizing the makespan of a single batch-processing machine with non-identical job sizes. The particles are first divided into three groups based on the fitness of the particle to propose a new method for updating location and velocity. Then an adaptive strategy for varying parameters of PSO based on cloud model theory is introduced and different groups adopted different inertia weight generating methods, which does not only improve the convergence speed, but also maintain the diversity of the population. The global search performance of this adaptive algorithm is validated by the results of the comparative experiments．