融合MCMC 方法的差分进化算法 在水文模型参数优选中的应用

传统差分进化算法在优选水文模型参数时容易出现/ 早熟收敛0问题, 基于马尔可夫链蒙特卡罗方法的差分进 化算法) ) ) DREAM 算法, 对嘉陵江流域降雨径流模型的参数优选问题进行了分析。结果发现, DREAM 算法融合 了自适应Metro po lis 方法的优点, 能有效克服/ 早熟收敛0问题, 适用于推求先验信息较少的复杂水文模型参数后验 分布。 Premature conver gence pro blem ex ists in the parameter o ptimization of hydro log ical mo del using the t raditio nal differ2 ential evo lution. In this paper, differ ential evo lutio n adaptive metro po lis algo rithm( DREAM) w as pro po sed, w hich combines the advantag es o f differential ev olut ion algo rithm and Mar ko v Chain M onte Carlo( MCMC) sampler , and applied in the parameter optimizatio n o f CMD23PAR hydro log ic model in the Jialing River Basin. The results show ed that DREAM has the advantag es of self2adaptiv e Metro po lis method, can effectiv ely o vercome the pro blem o f premature converg ence, and is capable to infer the pos2 ter io r distributio n o f model parameters w hich is lack of prior info rmation. 国家自然科学基金( 50809058)