Life Prediction Of SCR Denitration Catalyst In Coal-Fired Power Plants

Nitrogen oxides are a group of the pollutants that seriously affect the ecological environment.With increasingly strict environmental protection policies in China,coal-fired power plants are paying more and more attention to the control of nitrogen oxides in gas emission.Selective reduction technology(SCR)is the most important nitrogen oxides emission control technology in coal-fired power plants worldwide,of which the core is SCR catalyst.The determining factor on operation of SCR system is the catalyst performance.Therefore,accurate prediction on the operating status and remaining life of SCR catalyst is fundamental for the safe and economic operation of coal-fired power plants.Currently,SCR catalyst life is mainly analyzed and predicted by exponential model and single factor inactivation kinetic model.However,due to the complex operation environment of SCR catalyst and variable flue gas conditions,deactivation of catalyst is the result of multiple physical and chemical factors.Therefore,the principle of catalyst deactivation cannot be accurately predicted by traditional physical or mathematical formulas.Based on the big data of power plants,data that is pre-processed in the perspective of data mining are employed to study the variation of catalyst activity through single factor inactivation model and multi-factor inactivation model.In the single factor inactivation model study,a fitting curve and a gray prediction model are established to analyze the effect of running time on catalyst activity.In the multi-factor inactivation model study,the catalyst activity is predicted by BP neural network model and grey neural network model.Through comparing the examples,it is found that the pre-processed data satisfies the isochronal characteristics.At this time,the optimized direct output model of the gray neural network shows higher prediction accuracy.For power plants with large fluctuations in flue gas parameters,the optimization is conducted considering both data pre-processing method and prediction model.In data pre-processing,the amount of flue gas is determined as the standard for data screening.Besides,aiming at the structure defects of BP neural network,the neural network modeling research of genetic algorithm optimization is carried out.The results show that data screening can avoid the influence of fluctuations in flue gas parameters on the prediction accuracy.The screened data no longer has the isochronal characteristics.The genetic algorithm optimization network(GABP)shows the best performance of prediction.