This article is devoted to the methods of numerical and symbolical methods of calculations, which are based on perturbation theory and are implemented in terms of matrix formalism. These methods were used to solve various problems of modeling and optimization in accelerator physics. The basic theoretical principles are verified using methods to neural network technologies, machine learning and others.