Information-extreme machine learning of the control system over the power unit of a thermal power main line

Abstract

The study considers a method of deep machine learning of a decision-making support system of control of a power unit of a thermal power plant. We developed a method within the framework of information-extreme intelligent technology, it is based on maximization of informational capacity of a control system in the process of machine learning. We developed categorical models of information-extreme machine learning with optimization of control tolerances to recognition attributes and levels of selection of coordinates of averaged binary vector-realizations of recognition classes. We considered a modified Kullback information criterion as a criterion for optimization of learning parameters. We implemented algorithms of machine learning with polymodal and unimodal decisive rules. We formed a learning matrix based on archival data of the operation of Shostka thermal and power plant. The results of physical modeling showed that the use of polymodal decisive rules does not provide a high functional efficiency of machine learning. We ordered the alphabet of recognition classes to the magnitude of deviation of a functional state of the technological process from the standard regime for the application of unimodal decisive rules. At the same time, we constructed unimodal decisive rules according to geometric parameters of hyper-spherical containers of recognition classes х by the enclosed structure. We proved experimentally that the use of the unimodal classifier gives possibility to construct decisive rules, which error-free by a learning matrix. The obtained results give possibility to provide high functional efficiency of machine learning of control systems of technological processes whose classes of recognition intersect substantially in a space of attributes.