A. N. Vasilyev, V. P. Osipov, D. A. Tarkhov

A unified approach to the solution to a broad spectrum of PDE problems is offered in the paper, this approach is worked out in terms of ANN - artificial neural network - methodology. The possibility of various methods (finite differences, finite elements and neural networks) consideration from one point of view is shown. Stated methodology allows us to solve both direct and inverse problems uniformly, to take into consideration replenished experimental data, to build a hierarchy of tunable models, etc.