Neural differentiation in modeling

Abstract

The paper deals with a promising approach of modeling the real life systems, characterized with sets Of measured/discrete data, by replacing them with analytical functions framework. The article is focused on neural network approximation of functional expressions. As an analyzed system a dynamic flight model has been chosen due to the necessity of considering several classes of large sets of aerodynamic lift, drag, speed, force, balance and mass data to get a comparable mock-up response. Handling such type of model is naturally a huge computation time demanding process. Being able to substitute it with analytical functions system presenting a coincident behaviour could dramatically improve computation time at all aspects of utilization (UAV/UAS, autopilot systems, flight simulators, real lime control & stability response determination, etc.). Therefore first steps how to obtain analytical function are shown here. In this paper, sample case parameters were used to produce data that were then fitted with an exact function obtained from feedforward neural network