Dynamic model of waste heat boiler based on recurrent neural network

Abstract

The work is devoted to the development of a dynamic model of a waste heat boiler based on a recurrent neural network. The developed model can be used to create computer simulators for gas turbine plant operators, technologists and operating personnel. The object of modeling is presented as a complex thermodynamic system. The dynamic processes taking place inside the boiler are non-linear and interconnected. Changes in the technological parameters of the exhaust gases occur in ranges that do not allow to obtain an acceptable quality of the linearized model. Due of the difficulty of creating a mathematical description that takes into account the operation of the installation in different modes, recurrent neural networks were chosen to implement the simulation task. Based on the recurrent neural network, a dynamic model was synthesized that describes the change in the technological parameters of the waste heat boiler in the Power boost, Rated Load, Power reduction operating modes. The model output is the temperature of the network water behind the boiler. The created model takes into account the change in the water flow through the boiler, the change in the inlet water temperature, the increase and decrease in the temperature and pressure of the exhaust gas at the inlet of the waste heat boiler. In the formation of training and test samples for the neural network, archival trends obtained during the operation of the waste heat boiler were used. The article provides experimental data, a description of the stages of the synthesis of a neural network model, structural and graphic schemes, simulation results with explanations.