Traditional frequency analysis methods, such as Pearson type III is usually based on hypothesis of probability distribution subjectively, and meanwhile, it brings larger uncertainty in the parameters estimation as well due to insufficient samples. Aiming at the shortcomings of traditional method, we proposed a method which maximizes the entropy of probability distribution based on the parameters optimization by accelerating genetic algorithm (AGA), called MEGA. Frequency analysis study of the disaster rate series from 1950 to 1990 in Heilongjiang Province shows that, the probability distribution functions obtained under the restriction of 3-order and 6-order moment can truly reflect the implicit risk level information of historical disaster samples. Comparing to the Pearson type III method and information diffusion method, results of MEGA-3 and MEGA -6 are well in accordance with the reality, and avoid subjective information in the analysis process. Annual expectation disaster rates by MEGA-3 and MEGA-6 are 2.71% and 2.61% respectively, which indicate that the flood of Heilongjiang Province is under low risk level. MEGA methods proposed in this paper have a believable and reasonable result, and they will bring a new study way into flood risk analysis.
