Research On Numerical Methods Based On Evolution Computation

With the rapid development of electronic computers and the widespread applicati- on, the scientific computing has surpassed and replaced the experimental technique gr -adually in the multitudinous domains. People have realized increasingly that the scie- ntific computing was the third method of the scientific research and was more and mo -re important. The computational method is a mathematics branch, whose object of st- udy is to solve the numerical methods and the related theories of each mathematics qu -estion using the computer. The content includes numerical solution of nonlinear equa -tion (group) function, digital approximation (interpolation and fitting), numerical int -egration and numerical differentiation and so on. Particularly, with the appearance of the new computational intelligence technology of late years, the research about nume- rical computation suited for computational intelligence method is becoming extreme- ly urgent and essential. Based on these, in the article we will use the evolution algori- thms of the computational intelligence technology to study the related questions of tr- aditional numerical computational methods. So far, there have been many problems in traditional numerical computational methods. For example, traditional algorithms hav -e sensitive issue of the initial point in computing roots of algebra equations, which c- an affect the character of solution directly when the initial point is not suitable. Tradi- tional algorithm divides approximately two steps in the date fitting. Namely it determ -ines the structure of the fitting equation firstly, then uses least square method to solu- te model parameter, but concreting the structure of the fitting function (high dimensio -nal function) is difficult actually. Traditional numerical integration methods are very difficult to obtain high accuracy and low complexity and so on.In view of questions about traditional numerical methods, prime task of this artic- le uses the characteristics of evolution algorithms, such as autoadaptive search, global convergence and robustness and so on, to solve problems of traditional numerical met -hods. This paper proposed numerical methods based on the evolution computations mainly including of solving the real roots of the random polynomial, the data fitting and the numerical integration and so on, which have the important theory value and the practical application background in regard to improving tradition numerical met- hods.