基于改进自适应遗传算法的新安江模型参数率定研究

Research on the Optimization of Xin′anjiang Model Parameters Based on Improved Adaptive Genetic Algorithm

模型参数率定是提高水文模型模拟效果的重要手段,通过研究一种改进的自适应遗传算法(IAGA)对新安江模型参数进行优化率定,解决传统遗传算法初始种群质量不高、容易早熟收敛、局部搜索能力差等问题。该算法利用混沌变量遍历性特点,随机生成初始种群并选优,提高初始种群的个体质量;针对交叉与变异的进化过程,设计了反映种群离散程度的种群目标函数离散系数,利用该系数构建了自适应调整交叉与变异概率算子,防止遗传算法过早收敛;依托环形交叉算子,提高算法全局搜索能力;采用自适应非均匀变异算子,实时优化算法的局部搜索能力,避免陷入局部最优。将自适应遗传算法、传统遗传算法(GA)和自适应遗传算法(AGA)应用于秦淮河流域新安江模型的参数率定,并从率定的收敛性、耗时、稳定性和效果方面进行算法的性能比较,结果表明:IAGA算法具有更优的寻优能力,更好的收敛结果,更高的稳定性和精度,场次洪水的模拟效果优于GA算法和AGA算法,率定期与验证期确定性系数(R2)均在0.85以上,纳什效率系数(NSE)均在0.8以上,总体达到了水文预报的乙级标准。结果表明采用上述的综合手段改进传统遗传算法是可行的,改进后的IAGA算法具有良好的应用前景,为新安江模型的自动率定提供了一种有效的途径。

Model parameter calibration is an important way to improve the simulation effect of hydrological models.In order to solve the prob⁃lems of low initial population quality,prematurity and poor local search ability of traditional genetic algorithm,this paper proposes an im⁃proved adaptive genetic algorithm(IAGA)to optimize the parameters of Xinʹanjiang model.The Ergodic characteristics of chaos variables are used to randomly generate the initial population and select the best,so as to improve the individual quality of the initial population.Aiming at the evolutionary process of crossover and mutation,this paper designs the discrete coefficient of population objective function to reflect the degree of population dispersion.By using this coefficient,the adaptive adjustment crossover and mutation probability operators are construct⁃ed to prevent premature convergence of genetic algorithm.Based on the ring crossover operator,the global search ability of the algorithm is improved.The adaptive non-uniform mutation operator is used to optimize the local search ability of the algorithm in real time and avoid fall⁃ing into local optimum.The IAGA algorithm,traditional genetic algorithm(GA)and adaptive genetic algorithm(AGA)are applied to the pa⁃rameter calibration of the Xin′anjiang model in the Qinhuai River Basin,and the performances are compared from the aspects of conver⁃gence,time-consuming,stability and effect.The results show that the IAGA algorithm has more excellent optimization ability,better conver⁃gence results,higher stability and accuracy.The flood simulation results are better than the GA algorithm and the AGA algorithm.During the calibration period and the verification period,the deterministic coefficients are higher than 0.85,and Nash-Sutcliffe efficiency coefficient are higher than 0.8,which generally meets the second-level standard of hydrological forecasting.The results show that it is feasible to im⁃prove the traditional genetic algorithm by using the above comprehensive means,and the improved IAGA algorithm has a good application prospect,which provides an effective way for the automatic calibration of Xin′anjiang model.