基于改进差分进化算法的水文模型参数多目标优选研究

在差分进化算法的基础上, 受马尔可夫链蒙特卡罗方法的启发, 建立了differential evolution adaptive metropolis (DREAM)算法. DREAM 算法融合了马尔可夫链蒙特卡罗方法和差分进化算法的优势, 较好地解决了马尔可夫链蒙特卡罗方法中搜索步长的恰当取值以及搜索方向的准确定位问题, 并能有效解决差分进化算法的群体多样性和收敛速度问题. 在 DREAM 算法基础上, 引入多目标优化思想, 提出了一种基于改进适应度分配策略和外部存档方案的多目标 DREAM 算法, 并应用于岷江流域 CMD-3PAR 降雨-径流模型参数优选研究. 结果表明: 多目标DREAM算法能够找到一组范围宽广、分布均匀且数量充足的 Pareto 最优解供决策者评价优选.

Study on modified multi-objective differential evolution algorithm for parameter optimization of hydrologic model

A novel differential evolution adaptive metropolis algorithm (DREAM) is presented, which combines the advantages of differential evolution algorithm and Markov chain Monte Carlo (MCMC) sampler. DREAM solves an important problem in MCMC, namely that of choosing an appropriate scale and orientation for the jumping distribution. Meanwhile, it can make a good trade-off between population diversity and convergence for differential evolution algorithm. Moreover, multi-objective DREAM is proposed based on the modified fitness assignment and external archive strategy, which is applied in parameter optimization of CMD-3PAR hydrologic model in the Min River Basin. The results show that DREAM is capable to infer the posterior distribution of model parameters, and multi-objective differential evolution adaptive metropolis (MODREAM) is capable to generate a lot of non-dominated solutions with wide and uniform distribution for decision-makers.