This paper established a mathematical model of reactive power optimization. For the shortcomings of standard PSO that easily falling into local minima and poor convergence precision, this paper apply an improved particle swarm optimization algorithm to the power system reactive power optimization. The paper improved the formula of speed of PSO and introduced arctangent inertia weight and threshold to enhance the ability to find the global optimal solution. Simulation results of IEEE 30-bus system show that the improve particle swarm algorithm for reactive power optimization problem is feasible. Comparing with the results of PSO, we can find that improved particle swarm optimization algorithm improve the calculation accuracy in a certain extent.