Implementace rozhodování na bázi neuronových sítí v programovatelném automatu

Abstract

This thesis focuses on the implementation of neural network in SIMATIC S7-1500 programmable logic controller. First it goes through the general theory of neural network together with two models of neural network, which are the model of the feed forward multilayer neural network with an algorithm of back-propagation learning and the model of Hopfield network. In the theoretical part there is additionally described current possibilities of neural network implementation to programmable logic controllers. In the practical part the implementation of the both models of neural networks is performed and additionally the user’s application is created, which enables to automatically generate the created models of neural network. The practical part additionally focus on usage of the model of feed forward multilayer network for the control of non-linear system, when a feed forward and inverse model of system is created with the help of the neural network, when these models are used for control the system with the help of internal model and using direct inverse control.