Study On Boiler Combustion System Modeling And Steam Temperature Regulation Of 1000MW Ultra Supercritical Unit

Large capacity and high parameter coal-fired generating units are the appropriate path to realize "carbon peak and carbon neutralization" in China.The Double Tangential Boiler has been applied to millions of thermal power units,but there is also a problem of large steam temperature deviation.Therefore,this paper studies the boiler combustion system modeling and steam temperature regulation of1000 MW ultra supercritical unit.According to the characteristics of the unit,and considering the economy and safety,55 operating parameters such as load,feed water flow and flue gas oxygen content are theoretically selected as the input parameters of the combustion system LSTM model,and the sum of flue gas heat loss,solid incomplete combustion heat loss and flue gas temperature difference at the furnace outlet are selected as the output parameters of the model.Collect more than 210,000 on-site operation data of the unit from May 1,2021 to October 1,2021 for steady-state condition screening and condition division.In order to reduce the influence of disturbance on data information,the sliding window steady-state data processing method based on the relative standard deviation of samples is used to screen the required steady-state data.DBSCAN clustering algorithm is used,according to the characteristics of variable load,coal quality and climate during unit operation,the data are divided into 67 working conditions with load,coal quality and ambient temperature as the boundary.In order to simplify the model and improve the operation speed of the model,the input parameters are reduced.The input parameters are divided into operation parameters of combustion optimization and other variables.26 other variables are transformed into 10 comprehensive variables by p rincipal component analysis as part of the input of the model.The combustion system model is built by LSTM neural network.Through the experimental analysis of the model accuracy of different structural parameters,the double hidden layer model with time step of5 min,hidden layer node of 11,learning rate of 0.004 and Adam optimizer is finally adopted.Finally,particle swarm optimization algorithm is used to optimize the combustion operation parameters.Firstly,the optimal structural parameter of particle swarm optimization algorithm,namely learning factor,is determined,the learning factor is 1.5 and the inertia weight is 0.5;Aiming at the maximum thermal efficiency and the maximum deviation of flue gas temperature on both sides meeting the safety conditions,the operating parameters are optimized.The experimental research is carried out under the working conditions of 50% B RL and80% BRL.After the parameter adjustment,the sum of the two heat losses of the boiler is reduced by 0.56% and 0.29% respectively,and the flue gas temperature difference on both sides is reduced by 8.2℃ and 6.3℃ respectively.The results show the effectiveness of this method;The optimization test of the unit under all working conditions shows that the average heat loss under all working conditions is reduced by 0.45%,and the average flue gas temperatu re difference on both sides under all working conditions is reduced by 5.2 ℃.