多目标进化算法在新安江模型参数率定中的应用比较研究

Comparison between Multi-objective Evolutionary Algorithms for Calibration of Xinanjiang Model

为了克服传统优化技术在新安江模型参数率定中收敛慢、不稳定等缺点,近年来利用进化算法在水文模型的率定中得到越来越多的重视和发展,但对它们优选效果的比较讨论却非常少见。本文根据三水源新安江模型特点,以洪峰流量,峰现时间和洪水总量为优化目标,采用NSGA-Ⅱ、SPEA、PESA三种经典算法,通过对三种算法产生的非支配集进行算法评价,并比较了三种算法产生非支配集的收敛性与分布性,结果表明NSGA-Ⅱ在总体上最优。

Application of traditional methods to calibrate the Xinanjiang model will lead to disadvantages of slow convergence and instability.In order to overcome these disadvantages,the evolutionary algorithms towards the hydrological model calibration have been paid more and more attention,but the comparison of the algorithms is rarely discussed.According to the characteristics of the Xinanjiang model,the peak flow,peak time and flood volume were chosen as three criteria.The NSGA-II,SPEA and PESA were used for calibration and the results proved the effectiveness of the multi-objective evolutionary algorithms.The non-dominated set created by the algorithms was applied to test their performance of convergence and the distribution.The results indicate that the NSGA-II is better than the others.