Research On On-line Optimization Of Air Distribution System Of 1000MW Double Tangential Boiler

Ultra-supercritical power generation technology is the most importantly clean and efficient coal power technology in my country.A scientific and reasonable air distribution method has an important impact on the stability of boiler combustion,combustion efficiency,and nitrogen oxide emissions.In actual operation,the unit is also faced with variable load and coal quality,which puts forward higher requirements for operation.To provide operators with real-time operation guidance and organize good combustion in the furnace,this paper conducts an online optimization study on the air distribution system of a 1000 MW double tangential boiler.Double tangential combustion uses two relatively independent reverse tangent circles.By analyzing the influence of the air distribution method on the combustion process,determine the unit load,coal quality information,total air volume,primary and secondary air temperature,coal feed volume of each layer of coal mill,the pressure difference between secondary airbox and furnace,and burner 42 variables such as swing angle,secondary air door opening of each layer and burn-out air door opening are input influencing factors.Collected more than 880,000 pieces of actual data on site for one year,divided the working conditions according to load,ambient temperature,and coal quality information,and carried out data preprocessing.The steady-state data extraction method based on the time window of the sample standard deviation was used to eliminate non Steady-state data and over-temperature and over-limit data.In this paper,DNN deep neural network is used,and the target value formula for comprehensive optimization of air and powder is used as the output parameter to establish a boiler air distribution system model.Thr ough many experiments,the optimal parameter combination of the neural network is determined,that is,the number of hidden layers is 4,the number of neuron nodes is 20,the learning rate is0.02,and the optimizer is Adam.Through the training of 2700 gr oups of working conditions,a relatively complete model library of the air distribution system has been established.To guide the actual operation,according to the operating regulations,the operating parameters and operating range have been determined.Apply genetic algorithm and particle swarm algorithm to optimize online operation parameters.Firstly,the hyperparameter combination of the two algorithms is determined.For genetic algorithm,mutation rate Pm=0.01,crossover rate Pc=0.9;for particle swa rm algorithm,learning factor c1=c2=2,inertia factor w=0.4.Then,the optimization crossover test was performed on 2700 samples,and the comprehensive optimization value of wind and powder after optimization was reduced by an average of 0.25.Finally,under three typical loads(60%,80%,100%),on-site adjustment tests were carried out on the developed online guidance system.By optimizing the opening degree of each layer of the air door,the comprehensive optimization value of air powder has been reduced by 0.19,0.16,and 0.17,respectively,the NOx emission concentration has been reduced by 36.31 mg/Nm~3,48.63 mg/Nm~3,and 42.20 mg/Nm~3,and the boiler efficiency has been increased by 0.09%,0.06%,and 0.11%.