An improved particle swarm optimization algorithm combined with the natural selective mechanism is presented for distribution network reactive power optimization to reach the goal that the distributed generations are put into the distribution network with the minimum active power loss and ameliorate the node voltage. In this paper, the reactive power of DG generate to the system is considered as reactive power optimization control variables in distribution network,and then establish the mathematical model of the distribution network consists of the objective function, the equality constraint and inequality constraint of the power flow equations. The main idea of the intelligent algorithm is that the whole particle swarm is sorted according to the fitness value,the velocity and position of the worst half of the quantities are replaced by the best half of the quantities of the whole particle swarm, while the historical optimal value of each individual is conserved. The analysis of simulation in distribution network is carried out based on a mended IEEE33-node system, the conclusions verify the rapid global convergence and steadiness of the algorithm, and it can search the minimum active power loss with the fastest rate of convergence.