For wind power system, the uncertainty of wind leads to the instability of the particle' fitness and more pessimum particles, so it's difficult to quickly converge to the optimal value, which causes the large system voltage deviation and the sharp increase of network loss. A Dynamic Cloud Evolution Particle Swarm Optimization (DCEPSO) algorithm is proposed to realize the reactive power optimization of wind power system. Firstly, the minimum network loss is designed to be the optimization goal of reactive power optimization model of wind power system. Secondly, the DCEPSO algorithm is presented. According to the particle's fitness value, the algorithm selects excellent individuals to evolve, which reduces the proportion of pessimum particle and increases the search speed. Then through the cloud generator, excellent population evolved by excellent individuals tends to normal distribution, so as to improve the particle distribution. On this basis, according to the characteristics of the normal cloud distribution, dynamically changing speed can further improve the particle distribution and the search precision. Finally, the active network loss of wind power system is made to be the optimization goal to determine the capacity of compensation. The simulation results prove the effectiveness of the proposed method.