Study On Soft-sensing Theory Of NO_x Content In The Inlet Of SCR Reactor

The Selective Catalytic Reduction(SCR)technology has become the main choice to control NO_x emission in the coal-fired power plant because of its high denitrification rate,reliable technique,simple structure and small ammonia escape rate.The timely and accurate measurement of NO_x content in the inlet of SCR reactor is very important to regulate the amount of ammonia injection and also affects the emission of NO_x in the SCR reactor outlet.At present,the CEMS is widely used in thermal power plant to measure the content of NO_x,but the measurment is later than the actual value.In order to improve the measurement accuracy of NO_x content in the SCR reactor inlet,this paper uses the Least Square Support Vector Machine(LSSVM)to build the soft sensing model of NO_x content in the SCR reactor inlet,and realizes the accurate prediction of NO_x content.In this paper,the content of NO_x in the SCR reactor is the research object.Firstly,19 initial correlation variables are determined by analyzing the factors that relate to the content of nitrogen oxides.After processing the collected data,the 6 main auxiliary variables are identified by using the cross validity of partial least squares regression and the importance of variable projection,which aims to simplify the data and reduce the dimension.Based on the selected variables as input vector,a multi input single output least squares support vector machine soft sensor model is established by MATLAB programming.Then the particle swarm optimization algorithm is used to find the optimization of the least square support vector machine parameter penalty factor C and the radial basis kernel function width ?,which effectively overcomes the blindness of the grid search method to select the parameters,and the accuracy of the model is improved effictively.Finally,a dynamic LSSVM soft sensing model is established,and it is based on the real-time prediction error,adaptively modifies the model parameters.The old training samples are discareded,while the new data is added as the training samples to establish the LSSVM model,and the model can better predict the SCR reactor inlet NO_x content and correct the model on line..Using the LSSVM model to predict the SCR reactor entrance NO_x content effectively solves the lag problem of measuring the NO_x content,which is conductive to improving the accuracy of the amount of ammonia injection,reducing the ammonia escape,and reducing the SCR reactor outlet NO_x emissions.It is significant to the energy-saving emission in power plant.