This paper presents a new method based on the use of an optimization approach along with kernel least mean square (KLMS) algorithm for solving ordinary dif- ferential equations (ODEs). The new approach in com- parison with the other existing methods (such as numerical methods and the methods that are based on neural net- works) has more advantages such as simple implementa- tion, fast convergence, and also little error. In this paper, we use the ability of KLMS in prediction by applying an optimization method to predict the solution of ODE. The basic idea is that first a trial solution of the ODE is written by using the KLMS structure, and then by defining an error function and minimizing it via an optimization algorithm (in this paper, we used the quasi-Newton BFGS method), the parameters of KLMS are adjusted such that the trial solution satisfies the DE. After the optimization step, the achieved optimal parameters of the KLMS model are replaced in the trial solution. The accuracy of the method is illustrated by solving several problems.

Keywords