Research On Multi-optimization Of SCR System Based On Data Driven Algorithm

The deep peak regulation in coal-fired power plants affects the stable operation of SCR systems and causes frequent changes in SCR outlet NO_x emissions.The SCR denitrification system has the advantages of easy installation and high denitrification efficiency.The power plants widely utilize SCR systems to reduce NO_x emissions.Therefore achieving efficient and economic operation of SCR systems is one of the key issues faced by the coal-fired power plants today.Aiming at solving the problems of delay time calculation of relevant parameters,NO_x emission dynamic prediction modeling,and intelligent optimization adjustment of ammonia injection volume of SCR system,this paper constructs the NO_x emission dynamic prediction model and multi-objective optimization algorithm with the 1000 MW coal-fired unit denitrification system as the research object.The specific work is as follows:Firstly,given the complex denitration reaction mechanism and delay characteristics of SCR system,it is difficult to establish an accurate NO_x emission prediction model.This paper proposes a hybrid data-driven algorithm based on deep neural network for solving the modeling problem.This paper constructs the combined feature selection algorithm（MSN）based on CART,RF,XGBoost,and MIC to obtain the feature variables.With the average MIC between the reconstructed feature sequence and the SCR outlet NO_x emissions as the optimization objective,this paper utilizes the JAYA algorithm to solve this optimization problem.Reconstruct the modeling data according to the delay time obtained by the JAYA algorithm.This paper establishes the initial NO_x emission prediction model based on DNN,and proposes an error correction strategy to realize the early correction of the initial prediction error.Thus,a high-precision dynamic prediction model of the SCR outlet NO_x emissions is established.Secondly,the ammonia injection volume is selected as the optimization variable.The first objective function is the absolute deviation values between the SCR outlet NO_x emission and the desired value within the optimization time window.The second objective function is the cost of ammonia injection.In addition,the constraint on the range of variation of the ammonia injection volume is added.This paper uses the NSGA-II algorithm to solve the above multi-objective optimization problem and obtain the Pareto front.It ensures the NO_x emission is close to the desired value while considering the cost of ammonia injection.This algorithm guides the optimal setting of ammonia injection volume under the actual operating conditions.Finally,the software platform of ammonia injection volume multi-objective optimization system is developed based on C# and SQL server database.The SQL server database accomplishes the data storage and transmission,etc.The software platform collects functional modules such as user login,data management,modeling and prediction,multi-objective optimization,result display,etc.The user can query operational data and set parameters for modeling and optimization.The software platform completes the NO_x emission prediction and ammonia injection volume multi-objective optimization.Based on the historical operating data of the power plant,the hybrid data-driven algorithm constructed in this paper can accurately predict NO_x emissions in advance.The mean absolute percentage error（MAPE）of the prediction results is 3.87%.The established multi-objective optimization model of ammonia injection quantity provides a suggestion of the optimal setting values of ammonia injection volume.