Synthesis of Control Systems for Dynamic Pro-cesses in a Boiler Using a Neural Network of Direct Propagation

Abstract

The article solves an urgent scientific and technical problem of reducing the cost of fuel for combustion in steam boilers by automating the process of adjusting the gas-air ratio, which will significantly reduce the cost of certified fuel and rationally use emissions from industrial installations, such as industrial furnaces, oil refineries, etc. The development of an air flow regulator for natural and blast-furnace gas burners will save valuable natural resources and expand the range of use of emissions as secondary raw materials. Goal. Development on the basis of a neural network of direct distribution of a control system for the supply of natural and blast-furnace gas to the burners of boilers operating on one steam line. Methodology. The synthesis of a neuro-regulator for air supply during the combustion of non-certified gases in a boiler for the generation of thermal energy or electricity was carried out by means of Simulink in the MATLAB system. In order to implement the regulation of the combustion air supply, a neuro-regulator with foresight was used, the operation of which is based on the principle of a receding horizon, according to which the neural network model of the controlled process foresees the reaction of the control object at a certain time interval in the future. The procedure for the synthesis of a neuro-regulator for the supply of air for combustion is given. The values of the parameters that provide the specified indicators of the quality of the dynamic process control system functioning are established. The model is trained offline using the available empirical information regarding the process of gas combustion in the boiler, obtained on the basis of production experience. The settings of the resulting neural network are carried out according to the test data of a real object. Results. The universal possibilities of approximation of multilayer artificial neural networks of direct propagation made it possible to solve the problem of identifying, designing and modeling nonlinear control systems. The implementation of the developed neuro-regulator with the anticipation of the combustion air supply allows the control system to adapt to changes between theoretical and actual indicators and take into account and compensate for the variable parameters of the air supply to the burners. Conclusions. A feed-forward neural network has been successfully implemented in the synthesis of control systems for dynamic processes of regulating the air flow to natural and blast-furnace gas burners.