In this paper, an improved GA was proposed to minimum the disadvantages of classic GA. This modified GA improved the crossover and mutation strategy by using of sort program to arrange chromosome from big to small. The crossover probability and mutation probability were decided by the numbers of the order. Samples were chosen to take part in interbreeding. The convergence speed and results can be improved in this way. Moreover, premature convergence and local convergence were avoided at the same time. The modified GA was implemented by an optimization program compiled in visual C++ language. It has been successfully applied in the reactive power optimization of a Hydropower station. The results showed the feasibility and validity of this modified genetic algorithm.