Sensitivity Analysis For Control Parameters Of Hybrid And Standard PSO Algorithms: Application Via A Rainfall-runoff Model Calibration

In the phases of hydrological model calibration, the impact of control parameters of optimization algorithms on the related fitness (cost) function (reaching the global solution correctly and expeditiously) is rather essential. These control parameters’ dynamic structure and interactions can force the quantification of the mentioned influence. In recent years, both the decomposition of the uncertainties of variables representing any process and searching parameter sensitivities were conducted by means of variance analysis termed as ANOVA. In the study, the success of a hybrid PSO (HPSO) algorithm, which was composed of particle swarm optimization (PSO) and a gradient-type algorithm, was shown compared to a standard PSO for a monthly rainfall-runoff model example. In addition to the measures such as reaching stable results under each run and fast convergence, individual and interactive effects of parameters of employed algorithms on simulations were also investigated by ANOVA.  In the application, different combinations were assigned for the coefficients c1 and c2 that were defined in both PSO and HPSO, and then, algorithms were run 10 times in Acisu sub-basin of Gediz Basin. The evaluations have shown that the parameters’ uncertainties due to their individual behaviors and their interactions with each other are fairly reduced in HPSO implementation. In this sense, the fact that HPSO is not excessively sensitive to its control parameters has made it a preferred choice in the hydrological model calibration process.