The traditional particle swarm optimization (PSO) algorithm has some shortcomings, e.g. not easy to converge to the true Pareto front, and the poor diversity of the solutions. This paper presents a multi objective PSO algorithm based on adaptive network and dynamic crowding distance. For the case that the number of external population exceeds the population size, the proposed algorithm can divide the objective function space into the grids with the same interval, and then, estimate the density of particles by computing the number of particles in each grid so as to limit the size of the external archive. Moreover, the variance of particle is introduced and a dynamic crowding distance based algorithm is designed. That can avoid the problem that the distribution of solution become worse due to eliminating all individuals with small crowded distance in one time. Hence, the diversity of the solutions can be improved. Both the experiment on function optimization and the application in reconciling refined petroleum products validate the effectiveness of the proposed algorithm.