Emission Concentration Prediction Of Nitrogen Oxides And Sulfides In Thermal Power Plants

According to the latest standard of air pollutants for thermal power plants issued in 2014,all generators must have a NOx and sulfide emission concentration of less than 50 mg/Nm~3.This is a new challenge for thermal power plants.Since the coal quality of coal-fired power plants is not uniform,the amount of pollutants generated during combustion is not uniform.Moreover,the amount of pollutants will also vary with the load command and the changes in coal supply,air supply and other complex factors.In addition,most of the circulating fluidized beds in China use the SCR flue gas denitration system.As for the desulfurization,the limestone-gypsum wet desulfurization technology,namely WFGD technology,is adopted.However,both SCR and WFGD are large inertia and large delay systems,which brings great trouble to the control of pollutants at the flue gas outlet.At present,the method of controlling nitrogen oxides and sulfides at the flue gas outlet is mainly based on PID control method,and the opening degree of the ammonia spray valve and the opening degree of the limestone slurry valve are controlled by PID.Such control means,because of the lack of predictive ability to the target,it is difficult to achieve precise control.In order to make the emission concentration not exceed the standard,the power plants have to reduce the predetermined emission concentration,however,that can increase the flow rate of limestone slurry and increase the amount of ammoniasprayed.So this can lead to a large increase in the cost of desulfurization and denitrification,resulting in material waste,increasing ammonia escaping,and the PH value of the slurry in the slurry pool can be larger,thereby increasing the cost of desulfurization and denitrification,reducing the economic benefits of power plants.This paper focuses on the predicting the concentration of nitrogen oxides and sulfides in thermal power plants.By studying the sliding average model,least squares support vector machine,BP neural network,genetic algorithm,etc.The predicting models of nitrogen oxide and sulfide emission are modeled respectively by these methods.In addition,this paper also focuses on the combination of different algorithms and the actual operation of the power plant,so that the algorithms and the practice can be correspond as much as possible.Finally,the simulation results of actual operation data show that the proposed models are feasible.This paper has done the following work In terms of denitrification,Firstly,the correlation coefficient method is used to classify many variables affecting NOx generation and six main variables are selected.Secondly,the correlation coefficient method is used to determine the time-delay parameters of the three input variables in the prediction model.Then BP neural network method is used to select the three most Influential variables with the greatest influence as the input variables in the prediction model.And the Enumeration is used to ifentify the structure.Next,the least squares method is used toidentify the coefficient of the moving average model.Finally,the actual data from the power generation process of a power plant is used to verify the effectiveness of the proposed prediction method.In terms of desulfurization,the work of this paper is as follows:Firstly,Select the data set for the study.Because the desulfurization system has more influence factors than the denitration system,the relationship between the two variables in the data is more difficult to distinguish,and the system time lag is more difficult to determine.Therefore,after we draw the data curve,we determine a certain data set with a certain time lag to do research.Secondly,the predicted variables of the least squares support vector machine model are optimized.Previous work has focused on predicting the concentration of sulfur dioxide at the outlet.In order to make the least squares support vector machine model have a better performance,this paper shifts the data according to the time delay to eliminate the time lag effect.Next,the concentration of sulfur dioxide at the inlet of the flue gas is made to be different from the concentration of sulfur dioxide at the outlet of the flue gas.The difference is used as the predictive factor of the model.Thridly,the data with the same time lag is classified and use the correlation coefficient method to select representative variables.Then,data is translated according to time delay and then imported into the LSSVM model as a new data set for modeling.Fourth,PSO is used to optimize the LSSVM model parameters to further improve the fitting accuracy.