In view of the critical shortcomings such as local convergence and low efficiency in univariate marginal
distribution algorithm(UMDA), a hybrid univariate marginal distribution algorithm(HUMDA) is proposed. In the proposed
method, a two-stage dynamic parameters control strategy is used to control the mean and variance parameters in order to
preserve the diversity of population at the beginning of algorithm and improve the local search capability of the algorithm at
the end of the execution. In addition, the chaotic search strategy is adopted to enhance the precision of solution and search
efficiency. The HUMDA algorithm is tested by the high-dimensional and multimodal functions. The test results show that,
the proposed algorithm has better global convergence ability and search accuracy, which is applied to the optimal operation
of a reservoir and better results are obtained.