DEVELOPMENT OF NEURAL SYSTEMS FOR MONITORING AND CONTROLLING EMISSION OF GAS-TRANSMISSION AND POWER GAS TURBINE UNITS

• Main Authors: Valeriy G. Avgustinovich, Tatiana A. Kuznetsova, Alexey D. Nugumanov
• Format: Article
• Language: Russian
• Published: Tomsk Polytechnic University 2019-08-01
• Series: Известия Томского политехнического университета: Инжиниринг георесурсов
• Subjects: gas-turbine unit; low-emission combustion; automatic control and monitoring system; artificial intelligence; neural network
• Online Access: http://izvestiya.tpu.ru/archive/article/view/2207/2030

The relevance of the research is caused by the need to solve the problem of real-time monitoring and automatic control of nitrogen and carbon oxides emissions during operation of gas turbine units in gas compressor units and next-generation power plants characterized by a low level of harmful substances generation. The main aim of the research is compliance with emission standards while ensuring the stability of combustion under the influence of external and internal factors based on the creation of robust control algorithms for low-emission combustion chambers of gas turbine gas-compressor units of compressor stations of main gas pipelines and power plants, including prediction of their environmental effects, based on artificial intelligence technologies. Object of the research is low-emission combustion chamber of gas turbine gas-compressor units of compressor stations of main gas pipelines and power plants. Methods: design procedure for multilayered artificial neural networks based on the Arnold–Kolmogorov–Hecht-Nielsen theorem; back propagation algorithm; methods of full-scale experiment for low-emission combustion chambers; simulation methods and model experiment in the MATLAB environment. Results. The main features of low-emission combustion chamber of gasturbine units are considered. Low-emission combustion chamber tendency to unstable operation on the one hand is caused by the proximity of the operating mode to the boundary of the «poor» blowout and on the other hand – by the proximity of the combustion vibration mode, noted as the main automatic control problem. The problem of emission automatic control as a minimization of the share of fuel consumption through the diffusion contour is formulated, taking into account the limitations on the stability of combustion when the external and internal factors are changed in a wide range. The solution of automatic control problem based on artificial intelligence technologies including a built-in mathematical model for harmful substances emission is substantiated. The authors have developed the algorithm for low-emission combustion chamber mathematical model design based on artificial neural networks, taking into account the significance of the influence factors. As an example of solving the problem, the neural network developed and the process of its learning based on the experimental data of the real low-emission combustion chamber are presented. The data array of a full-scale experiment was obtained for studying the characteristics of emissions of the nitrogen and carbon oxides (NOx and CO) during operation of the low-emission combustion chamber of industrial power plant (16 MW). The neural circuit simulating NOx emission and CO emission at the low-emission combustion chamber output was designed and trained on the basis of the obtained data. The simulation results in the MATLAB environment showed high accuracy of the developed model. The importance for the model accuracy of different factors is studied. It turned out that temperature and pressure parameters are the most important. The obtained results can be used in the fault-tolerant system design for automatic control of gas turbine units to improve their reliability and environmental attractiveness.