According to the feature of reactive power optimization in distribution network with distributed generation, a new kind of hybrid algorithm, which is composed of invasive weed optimization (IWO) and differential evolution (DE) is proposed. A group of initial feasible solutions are reproduced and spatial dispersed, some of individuals which are higher adaptation degree are selected by introducing competition mechanism when they reach the maximum allowable environment, and the best individual is retained through mutation, crossover and selection. The algorithm utilizes advantages of simple structure, less parameters and robustness of IWO, and overcomes the drawback of being trapped in local optimum and lower accuracy in combination with DE. Comparison with original IWO in IEEE 33-bus system simulation analysis, the results show that the algorithm has stronger global search capability and higher degree of convergence and can effectively reduce the power loss.