Calibration of rainfall-runoff models using the evolution strategy
2003
Fujihara, Y. (Kobe Univ. (Japan)) | Tanakamaru, H. | Hata, T. | Tada, A.
abstract The Evolution Strategy (ES) is one of the global-type optimization procedures. The algorithm of the ES is similar to the Genetic Algorithm (GA) in the respect that multipoint search and recombination used in the GA are also adopted. But, the main feature of the ES is that its principal search procedure is mutation, which is conducted by adding perturbation of normal random variable to decision variable, and the size of perturbation is updated by self-adaptation. The ES is applied to parameter estimation of the Tank model, which has 16 unknown parameters including 4 initial storage depths, and the search ability of the ES is verified by numerical experiments using error-free synthetic data. The ES is also applied to parameter estimation using historical data of the Eigenji, Osako and Syorenji Dam Basin. These synthetic and historical data studies show that the search ability of the ES and the Parallel ES is far superior to that of the Binary GA and the Real GA and the Parallel ES is equal or superior to that of the SCE-UA method, which is currently considered to be the most powerful optimization technique, in the consistency of estimated parameter values.
اظهر المزيد [+] اقل [-]