SummaryResults obtained by simulation usually depend on scalar or matrix mathematical representations of the studied process, on the tool of simulation and on identification an/or data acquirement errors.Generally for linear process, existing canonical matrix representations lead to simulation schemes which allow a correct description of the system working.In non-linear hypothesis, those canonical forms are more difficult to be implemented and moreover can imply incertainties which result of various drifts and propaaation of truncation errors varying with the mode of choosed simulation and the order of the studied system.We propose here an original matrix representation of non-linear systems, closed in a fiability view point to the modal representation used for linear processes and allowina a new simulation method.It consists on a specific decomposition of the n order non-linear system into n interconnected subsystems including (n-1) arbitrarily choosed linear and stable ones.Then the simulation scheme for this decomposition allows an excellent internal attenuation of errors inherent in the type of the used simulation. and also permits an easy to check convergence test of the used algorithms and simulation technics.A power plant simulation is presented to illustrate an extension of the proposed method to interconnected processes simulation.This application concerns the study of the connection on the same electrical network of two power plants.