Screening Analysis Of Influencing Factors Of Wet Desulfurization Efficiency Based On BP-Garson Combination Model

At present,China's largest emitters of atmospheric pollutants SO2 are still coal-fired power plants.The desulfurization technology with the highest utilization rate in China's thermal power coal-fired units is limestone/gypsum wet flue gas desulfurization?WFGD?.Desulfurization efficiency is used as an important indicator to measure the power plant desulfurization system.,Directly affect the amount of pollutant emissions,so clarifying the degree of influence of various parameter variables in the desulfurization system on it has important guiding significance for improving desulfurization efficiency,reducing pollutant emissions,and formulating fault detection plans.With the application of information technology in the desulfurization system,the desulfurization system has accumulated a large amount of historical operating data in actual operation.Data mining technology can use the historical operation data of the desulfurization system to discover the parameter variables for desulfurization under certain operating conditions.The weight of the efficiency influence,and then excavate the factors that affect the desulfurization efficiency under the current working conditions,so as to provide operation guidance and optimize the production process for the operator.In this paper,the limestone/gypsum wet desulfurization system is taken as the research object.The algorithm of combining BP neural network and Garson algorithm is used to study the 180-day data of 45 parameter variables adopted in the DCS system.Model training and testing.First,pre-process the 180-day historical operating data of the unit to eliminate inferior and overlapping information,retain valuable information,and increase the value of the data.Data preprocessing includes cleaning,integration,transformation,reduction,collinear,and dimensionality reduction processing.Based on the 0.82 set value of the autocorrelation R diagnostic method,20 parameters were detected to be collinear and deleted;the stepwise regression method was used to reduce the dimensionality of the collinearly processed parameters to simplify the data and facilitate modeling.Finally,17 parameters are retained as input for modeling.Secondly,according to the characteristics of the operation data of the desulfurization system,the incremental mining technology is combined with BP neural network,and it is applied to the weight of the desulfurization system operation parameters to influence the desulfurization efficiency.Accuracy and model weight matrix.At the same time,R,RMSE and MAPE are used as model evaluation criteria to verify the quality of the model and provide basic data support for the screening of desulfurization efficiency impact factors.Finally,the weight coefficients of the input variables for the desulfurization efficiency are calculated according to the Garson algorithm,and the variables with higher frequency and larger amplitude are determined by the relative change values of the weight coefficients of the input variables in the 13 groups of the same structure and parameter model.Combined with the mechanism analysis,it is determined that the variables selected by this model are the factors that affect the desulfurization efficiency under this working condition.The study found that the desulfurization efficiency in actual operation is mainly affected by the pH value of the gypsum slurry,the mixed flue gas O₂,the process water flow of the demister,the compensated tower liquid level,the demister pressure difference,and the filtrate tank fluid,which are basically the same as the operating experience.Consistent.At the same time,the method of incremental mining can solve the problem of updating the impact factors after the operation state of the desulfurization system changes.