Research On Thermal System Modeling Of Coal-fired Generating Unit Under Flexible Operation

Increasing the flexibility of coal power operation is an important means to improve the capacity of renewable energy consumption.However,the process of coal-fired power generation has typical characteristics of large inertia,large delay,nonlinear and multivariable coupling.The traditional thermodynamic system control is designed based on the linearization of specific operating conditions,which cannot guarantee the control performance of the unit under flexible operation conditions of wide load.In view of this,it is necessary to conduct further modeling research on thermal power systems under flexible operation of wide load,and lay the foundation for the development of advanced control strategies.In this paper,a long-term and short-term memory neural network(LSTM)is used to establish the dynamic prediction model of main steam temperature and reheat steam temperature.Furthermore,the modeling method with data and mechanism is studied,and the dynamic model of coordination system is established for a subcritical unit and an ultra-supercritical unit respectively.This paper lays a foundation for revealing the characteristics of the thermal system under the flexible operation of wide load and designing the advanced control strategy.Taking a 660MW coal-fired boiler as the research object and using finite impulse response filter(FIR)for data filtering Taking a 660MW coal-fired boiler as the research object,a multi-parameter collaborative prediction model of main steam temperature,reheat steam temperature,NO_x concentration and CO concentration and a multi-step prediction model of reheat steam temperature are established based on LSTM.The research shows that FIR filter can effectively remove noise and spikes.The effect of the multi-parameter collaborative prediction model is better,and the correlation coefficient r of the four outputs is greater than0.93.The multi-step prediction model of reheat steam temperature is excellent,which could accurately predict the change of reheated steam temperature within 2.5 minutes.The average value of the root mean square error(RMSE)of the 5-step prediction result is less than 0.52℃.At the same time,it is found that the multi-step prediction model is better than the multi-parameter collaborative prediction model for the prediction results of the main steam temperature at time t+1.Based on the mechanism analysis and simplified hypothesis,a dynamic modeling method with mechanism and data of the engine-furnace coordination system is proposed,which are constructed respectively for the subcritical and ultra-supercritical units.In view of the complex mechanism and high modeling difficulty of some links,support vector machine(SVM)is used for auxiliary modeling.Then,a massive of operational data from a real power plant is employed to determine the dynamic parameters based on particle swarm optimization algorithm.Finally,Open-loop simulation and close-loop simulation upon history operational data are conducted to verify the developed model.The research shows that the modeling method with mechanism and data has higher precision than the traditional method.The results demonstrate that the model is capable to catch the dynamic characteristics of the actual unit and can predict output changes with satisfactory accuracy.The maximum value of average relative error of subcritical unit and ultra-supercritical unit is 1.77%and 3.24%respectively.The model can be used to design and improve the coordinated control system.