The Algorithm of Parametric Optimization of Automated Systems with PWM Element that Incorporates Artificial Neural Network

Abstract

Most optimization algorithms require prior appointment of its parameters. Formed on the basis of the method Nelder- Mead for the neural network learning algorithm (NNLA) was no exception. In this article the task specification values of the coefficients of the neural network learning algorithm (NNLA) is solved for systems containing PWM element that is composed of an artificial neural network. For this the genetic algorithm is applied to the most appropriate in this case selection strategy - "elitism". In order to expand the scope of formed algorithm NNLA, including automatic control systems in that processes are quickly introduced integral criterion, that along with the most common criterion, having in its composition an error, use the least amount of NNLA algorithm iterations. Assessment "health" formula is shown after the convolution operation of such criteria. The main variants of the neural network are considered: based on the modulation characteristics; single-layer fully connected neural network; single-layer fully connected neural network with feedback. The results of the application of genetic algorithm are given for determining the coefficients of the NNLA that configures an automatic system to achieve the integral quality criteria minimum, with use of the aforementioned embodiments of neural networks and features five activation of the neuron network.