First of all, the discusser would like to thank the author for using parameter-setting-free harmony search (PSF-HS) to estimate the parameters of the nonlinear Muskingum model and would like to draw attention to some points and suggest a practical way for parameter estimation of the nonlinear Muskingum model. In the original paper, the available methods for the parameter estimation of the nonlinear Muskingum model are impliedly classified as two groups. This division consists of (1) mathematical techniques, and (2) phenomenon-mimicking algorithms. However, it is notable that the procedures such as the neuro-fuzzy approach (Chu 2009) that cannot be classified in these two groups can be utilized for the purpose of estimating the parameters of the nonlinear Muskingum model. The discusser is unanimous with the author that the methods of the former group have weaknesses of the complex derivative requirement and/or initial vector assumption of design parameters, and the trouble of the methods of the latter group is that these methods need the determination of the algorithm parameters. However, it is notable that the latter group is randomly searched for the optimal solution, and some uncertainties that may cause different solutions on different runs existed when using the phenomenon-mimicking algorithms to estimate the parameters of the nonlinear Muskingum model. In this discussion, an alternative way was proposed for the parameter estimation to avoid the sensitivity analyses of either algorithm parameters or initial values assumption of the hydrologic parameters. In this procedure during two stages the methods of the mathematical techniques and phenomenon-mimicking algorithms are used. In this way, at first stage the phenomenonmimicking algorithm is run, and after a few seconds in the middle of the operation, the model can be stopped. In the second stage, the values obtained for the hydrologic parameters from the phenomenon-mimicking algorithm are considered as the initial guess of the hydrologic parameters for the mathematical technique. In this condition, the sensitivity analyses of either algorithm parameters or initial values assumption are not necessary. Therefore, finding the optimal solution can be done in the shortest operating time. If the near-optimal result is achieved in the first run of the mathematical technique, it is possible that the optimal result is reached with one extra run of the mathematical technique at the second stage. If infeasible, divergent, or faraway-optimal results are achieved, the procedure must be repeated from the first stage, but these results do not happen with a high probability. For this procedure, two combinations of methods are offered. These combinations consist of (1) genetic algorithm (GA; Mohan 1997) and Nelder-Mead simplex algorithm (NMS; Barati 2011) as GANMS, and (2) evolutionary (EV; Premium Solver Platform 2010) and Broyden–Fletcher–Goldfarb–Shanno technique (BFGS; Geem 2006; Premium Solver Platform 2010) as EV-BFGS. The GA-NMS and EV-BFGS procedures can be modeled by the Optimization Tool add-in of MATLAB software (Yang et al. 2005) and by the Excel Solver function add-in of Microsoft Excel software (Premium Solver Platform 2010), respectively. Therefore, these two procedures are widely available for engineers, and hydrologists
Read full abstract