Modeling Of Denitrification System Based On Machine Learning

At present,SCR denitrification system is widely used as the main means of NOx control in thermal power units in China.It is of great significance for the economic and environmental protection operation of thermal power units to establish an accurate denitrification system model and grasp the operation characteristic of SCR system.In this paper,a 660 MW coal-fired unit is taken as the research object,and the SCR denitrification system is modeled by machine learning method.The main research contents are as follows:Firstly,the mechanism of catalytic reduction reaction is analysed from the point of view of kinetics,and the mechanism model of SCR denitrification system is established.In order to improve the accuracy of the mechanism model,an improved particle swarm optimization algorithm is proposed,which combines the non-linear weight updating strategy and multi-target tracking strategy,and then a method of identifying the parameters of the mechanism model from the actual operation data is proposed.An example of a 660 MW unit shows the effectiveness of the method.Secondly,an improved sparse least squares support vector regression modeling method is proposed,which eliminates the singular point samples from the data based on the traditional pruning strategy and optimizes the core parameters by using IPSO algorithm.The IS-LSSVR method improves the prediction performance of the model while obtaining the sparse characteristics.A prediction model of the amount of injecting ammonia for SCR denitrification system based on IS-LSSVR is proposed.The prediction accuracy and generalization ability of the model are verified by numerical example.Thirdly,a LSTM-based soft-sensing model of NOx concentration at chimney entrance is established.The model takes the time series data of relevant variables of SCR denitrification system as input,and uses LSTM neural network method to extract the characteristic information from the time series data,thus realizes the soft-sensing of NOx concentration at chimney entrance.According to the time-varying characteristic of SCR system,the model updating strategy is introduced to improve the adaptive ability of the model.The models are validated by the operation data on site.Then,the soft sensor model of NOx concentration at chimney inlet is used to correct the NOx concentration at SCR outlet.Based on the corrected the operation data on site,two methods of catalyst performance deterioration analysis are proposed.Firstly,based on the mechanism model of SCR denitrification system,the kinetic parameters of the mechanism model are identified by using the corrected operation data,and the change of the activity of the catalyst during the service period is analyzed from the point of view of the change of the kinetic parameters of SCR denitrification reaction.Secondly,the prediction model of the amount of injecting ammonia for SCR system based on IS-LSSVR is applied to characterize the change of catalyst activity by the change of ammonia consumption at the same standard operating condition during the period of unit operation.The degradation trend of catalyst performance is obtained by time-division and multi-model modeling method.Finally,relying on the platform of denitrification catalyst integration of a group company,the software of SCR denitrification system condition monitoring and catalyst performance deterioration analysis is developed.