НЕЙРОННОЕ УПРАВЛЕНИЕ СИЛОВОЙ ТУРБИНОЙ

Abstract

Industrial facilities that include a power turbine as an object of control, have a continuous nature of the technological process, are complex and may not have a mathematical description. In the absence of a mathematical description of the object and in the presence of uncertainty, it is possible to apply the theory of fuzzy sets, as a branch of mathematics, where fuzzy logic and artificial neural networks (ANNs) are the implementation tool. The most promising are the radially-basic ANN, which, unlike the multi-layer INS perceptron type, is rapidly trained. The modification of RBF-network as an adaptive neural regulator for control of nondeterministic linguistic object is considered in the article. The network has one input, one hidden layer and an output layer of linear neurons. Training of the network (regulator) is carried out in two stages. At the first stage, the clustering of the hidden layer of neurons and their output is performed. At the second stage, the calculation of the synapses of the linear neurons of the output layer is performed from the known values of the output of neurons of the inner layer. Calculation of synapses of linear neurons of the output layer is performed by solving a system of linear algebraic equations, which increases the speed of learning the network by an order of magnitude higher than with multilayer perceptrons. However, in the intermediate layer for radial elements, it is necessary to determine the position of their centers and the magnitudes of the Gaussian windows. The use of the RBF network as a neural regulator is an alternative solution to adaptive fuzzy controller for solving similar problems.