IWRA Proceedings

AbstractExperience suggests that any single-objective search, no matter how carefully chosen, is not able to identify a solution capable of satisfactorily model a phenomenon of interest. Use of a multi-objective approach can yet be justified by the nature of real world problems, which in general involve multiobjectives, most of the time conflicting objectives. An approach very often used in multicriteria optimization is the concept of Pareto dominance, which allows us to compare different solutions by using different objectives and to explore different characteristics of the observed data. This paper employs evolutionary algorithms inspired on honey -bee mating for single- (Honey-Bee Mating Optimization - HBMO) and multi-objectives (Multi-Objective Honey- Bee Mating Optimization - MOHBMO) in the minimization of test functions and calibration of watershed models. The single-objective version of this algorithm was first introduced by Haddad et al. (2006), while its multi-objective version was proposed by Barros (2007). As reference of their performances, the following algorithms were used: PSO (Particle Swarm Optimization), and its multi-objective version MOPSO (Multi-Objective Particle Swarm), SCEM-UA (Shuffled Complex Evolution Metropolis) and its multi-objective version MOSCEM-UA (Multi- Objective Shuffled Complex Evolution Metropolis). Well known theoretical functions were used to test the proposed algorithms. Real world applications on hydrologic model calibration were also carried out in order to test the relative performance of the algorithms. In the case of test functions, PSO and MOPSO algorithms presented a better performance for the function that has a singularity. In general, HBMO and MOHBMO performed better than PSO and MOPSO/MOSCEM-UA, respectively, in terms of finding the extreme values of the Pareto front and adequately representing the front itself. PSO and MOPSO presented the worst performance among all algorithms when the test function used was the one that introduces a bias in the objective function. These algorithms were also applied in a multi-objective calibration study of the HYMOD model for many streamflow gauges in the State of Ceará, Brazil. In general over the state, MOHBMO and MOSCEM-UA performed better than MOPSO. Discussion on the choice of calibration and validation periods and the effects of the use of different objective functions on the calibrated parameter values is also presented.