実数を遺伝子とした遺伝的アルゴリズムによるデータあてはめ

富士市 吉本, Toshinobu Harada, 真光 森山 et al. · Institutional Repositories DataBase (IRDB) · 2000

In order to obtain a good approximation for data fitting with a spline,frequently we have to deal with knots as variables.Then, the problem to be solved becomes a continuous nonlinear and multivariate optimization problem with many local optima. Therefore it is difficult to obtain the global optimum.In this paper we propose a new method for solving the problem by a genetic algorithm with real number genes.In this method we use knots themselves as genes and we do not convert the original continuous problem into a discrete combinatorial problem.Therefore influence of the errors caused by the discretization of knots is avoided and quasimultiple knots can be constructed.We search for the best model among candidate models by using Bayes information criterion BIC.Our method can determine appropriate number and locations of knots automatically and simultaneously without good initial locations of knots.We do not need any subjective parameters.Moreover we propose a simple algorithm for multiplying quasi-multiple knots.Our method can treat data not only with a smooth underlying function but also with an underlying function having discontinuous points and/or cusps.Numerical examples are given to show the performance of our method.

Cited by 0 works.

Authors