Research On Modeling Of Simultaneous Desulfuri-Zation And Denitrification By A Biological Method Removing NO_X And SO₂ From Flue Gas Based On BP Artificial Neural Network

As the main parts of atmospheric pollutant, NOx and SO2 can be harmful. The flue gas discharged by coal-fired power plants need to be soon solved as it contains NOx and SO2, and contribute to main parts of total emission either.Among the technologies of purifying NOx and SO2 from flue gas, Now bio-purification technology is becoming a researching hot-spot as advantages following:always working in mild condition, no needing high temperature and pressure; and devices can be easily handled while take less land; secondary products have little pollution.This research is to the situation while techniques of Bio-purifying NOx and SO2 lacks a precious and effective prediction model. Because A prediction model can make the biggest purification efficiency during studying stage, it can also make a lowest cost during practical application stage. Most of all only an efficient, safe, stable, good handling, energy saving, economy facility combining with computer science, automatic control technology, can be used in the the practical productive activities. Among these an accurate mathematical model is an necessary foundation and worth investigating.While the bio-purification phrase is so complex, some influence factors are hard to quantify, That make an accurate mathematical model so difficult to build. While the BP neuron network has powerful nonlinear approximation capability,it can predict without a clearly function relation of output and input while just adjust its own structure like weight and threshold.This research propose a project which apply the BP network as modeling tool, then build prediction models of bio-purify NOx and SO2 from flue gas.The main tasks include:modeling of BP neuron network,verifying the model to prove veracity,improving BP model to increase prediction of precision ratio. These works have been completed:(1) Establish parameters used in building BP neuron network: Desulfurization-tower uses NOx air inlet concentration, SO2 air inlet concentration,gas flow.circulating-fluid flow,circulating-fluid pH.temperature to predict NOx outlet concentration and SO2 outlet concentration respectively. Denitrification-tower uses NOx air inlet concentration, gas flow, circulating-fluid flow.circulating-fluid pH,temperature to predict NOx outlet concentration. BP model has a structure of 3 layers, input layer and hidden layer choose TANSIG function, output layer choose PURELIN function, epoch phrase choose TRAINLM algorithm.The numbers of input layer,hidden layer,output layer neuron are:Desulfurization-tower purify NOx model:6,6,1: Denitrification-tower purify NOx model:5,5,1;Desulfurization-tower purify SO2 model:6,6,1.simulation experiments show, desulfurization-tower purify NO* has a precision ratio of 86.13%,denitrification-tower purify NOx model has a precision ratio of 90.7%. Desulfurization-tower purify SO2 model has no meaning now.(2) Use genetic algorithm to improve BP neuron network. Make the model fast calculate velocity and more accurate prediction.Building a GA-BP model with parameter of population size 50, generation 100,crossover probability 0.95,mutation probability 0.08, and same BP neuron network structure. simulation experiments show, desulfurization-tower purify NOx has a precision ratio of 92.7%, denitrification-tower purify NOx model has a precision ratio of 96.8%.While the average prediction errors of GA-BP model of double tower are less than 10% finally, even denitrification-tower purify NOx model is smaller than 5%, so the prediction models are effective. As a creative point,this research successfully combine the BP neuron network with simultaneous desulfurization and denitrification by a biological method removing NOx and SO2 from flue gas, hopefully achieve the goals of increasing the device of removing ratio during investigation, and promoting this technique toward practical industry.