This paper applies the idea of mixing leapfrog particle swarm algorithm and selection crossover operation of genetic algorithm to particle swarm algorithm, and put forward an improved binary particle swarm optimization to solve the problem of distribution network reconfiguration. Through update dynamically adjust the inertia coefficient of speed formula of particle swarm optimization, so that the particles can dynamically change the global and local search capability with the number of updates in order to prevent the algorithm from premature and find the optimal solution. Finally, the example of typical IEEE33 node is simulated and compared with genetic algorithm, simulation results illustrate that this method can avoid the algorithm premature effectively, with the advantages of rapid convergence and good stability.