Research On The Neural Network Predictive Control In The Application Of Denitration System

Due to the increasing environmental problems,environmental protection has became a hot topic in today's society.Most coal-fired power plants use SCR flue gas denitration technology,and how to effectively control NOx emissions in flue gas has became a key problem.Considering that the object of SCR denitration system has the characteristics of time denaturation,large lag,large inertia and non-linearity,the denitration control system based on the conventional control strategy is often difficult to obtain good control effect.The main research work is as follows:1?For 600 MW supercritical unit,a dynamic characteristic testin 350 MW,460 MW,550 MW three load-points separately was carried on with the collected experimental data,second-order model of large damping system identification method,control object of SCR denitration system in 350 MW,460 MW,550 MW three load-points was set up.At last the function model was transfered,the accuracy of the model was verified.2?Considering the SCR denitration system object is a big lag of the inertial system,On the basis of improving Smith's estimation compensation control strategy,A denitration control system based on cascade Smith estimation compensation technology was designed.The control system overcomes the high requirement of the traditional Smith estimation compensation control to the model precision,and has good robustness,rapidity and stability.The simulation results of the designed denitration control system are compared with the simulation results of traditional PID control,cascade PID control and conventional Smith estimation compensation control.The results show that the improved cascade Smith estimation compensation control system has better control effect.3?Considering that the controlled object of SCR denitration system is strongly nonlinear,firstly,the related technologies of multi-layer perceptron neural network and generalized predictive control are introduced.On this basis,an adaptive learning algorithm was uesd,recursive least square method to design a predictive controller based on neural network model,and further design the denitration control system based on neural network predictive controller.This paper put the denitration nonlinear systems based on Hammerstein model as simulation object,has carried on the forecast based on neural network control system of denitration simulation test,and with the improvement of cascade Smith prediction control system were compared with the simulation results of the linear predictive control system.The simulation results show that the denitration control system based on neural network predictive control has better control effect for denitration system with strong nonlinear characteristics.4?An improved particle swarm optimization(pso)algorithm based on gaussian distribution was put forward,through the standard function to test the proposed algorithm has the characteristics of the algorithm is simple rapid convergence,and overcome the premature convergence problem to some extent,the algorithm used in neural network predictive controller parameters optimization,solved the controller parameter setting difficult problem.