タンクモデル定数の最適化手法の改良

抄録

Operational comparisons are presented using the optimization techniques to identify the parameters in Sugawara' s Tank model which has been widely employed for long-term runoff analysis. Themathematical optimization techniques tested herein are both the Powell and Newton methods. Computational burden is too stringent in the application of the Powell method to parameter optimization in the Tank model, because a great number of runoff computations are required. The efficiency of optimization performance resulting from use of the Newton method clearly depends on how effectively the sensitivity coefficients can be derived. An important feature of the proposed approach is the theoretical derivation of sensitivity coefficients which can directly be used in the optimization scheme of the Newton method.
A vector differential equation in terms of storages in the four tanks is numerically integrated using the transition matrix which is computed by expanding the matrix exponential. The elements of this transition matrix play a significant role in eliminating additional computations involved in the solution of sensitivity coefficients. On the basis of the simulation results, the Newton method coupled with sensitivity coefficients appears to have potential performance advantages for optimizing the parameters in the Tank model.