Нейросетевой эмулятор нелинейной модели перевернутого маятника на подвижном основании

Abstract

This article demonstrates the neural network emulator design process of the plant model - inverted pendulum on a movable base nonlinear model. A feature of this plant model is the unequal number of input and output channels. The design of this emulator is necessary in the future to train a neural network controller by the method of back-passing an error through a direct neuroemulator. A distinctive feature of the demonstrated procedure for the neuroemulator design is the deterministic choice of architecture and initialization of the neural network weighting coefficients. The data on the choice of architecture and the values of the initialized weighting coefficients are based on information about the non-linear plant model. The plant model nonlinear parameters are taken into account due to the introduction of blocks selected for their approximation into the neural network emulator. As a result, a neural network capable of simulating the behavior of the plant at the required control range is obtained.