Research On Let Characteristic Of Boiler Based On Artificial Neural Networks And Genetic Algorithms

With the developing demand of environmental protection, thermal power generators are being confronted with two requirements to reduce its operation costs and to lower its emissions, especially NOx emission, which is one of the main pollutants released from coal-fired boilers. Boiler is a multivariable system; it is hard to figure out the emission property with a simple equation, because it is affected by many factors. Some experiments can roughly estemate the NOx emission, but it needs too many experiments, and it is difficult to analyze the experimental data, because it only represents the fixed condtion, the results without any generalization. The artificial neural network setting up an important characteristic of the mold is the black box of characteristic of its importation exportation, if treat boiler as a black box, the certain importation then is by all means to should certain exportation, so the application artificial neural network exhausts characteristic to set up a mold to the boiler.This text discussed influence the coal burning boiler NOx to exhaust in analyzing the coal burning boiler NOx bornly with breaking the foundation with mechanism of each factor.Use boiler hot test data, adopt three layer BP NN compose construct model of boiler let characteristic. Direct routine bp algorithm learning inefficiency, rapidity of convergence slow wait vice, adopt lot study, append momenta term, adaptive learning coefficient wait measure held over improve on it. Through the medium of boiler measured data demonstration, the BP NN is to the NOx, smoke temperature, fly ash to contain the carbon, row gas oxygen quantitative relatively prediction errors parting for 0.49%,-1.954%, 8.14%,-5.134%.Judge from the result of the examination, the estimate result of the NOx,the row smoke temperature nears to very much to measure a value actually.Direct initial weight cost and threshold versus cellular rapidity of convergence and error precision impact in favor of this one issue, adopt reis encode genetic algorithm priority nerve network weight value optimize, then adopt touch BP NN proceed optimize. The global search ability strong and BP NN local search ability strong point of the complex utilization genetic algorithm. Design and carry out to combine according to the heredity calculate way and the BP network of the boiler let characteristic network model, the GA BP network is to the NOx,smoke temperature and fly the ash contains carbon,the row smoke oxygen quantity of the opposite estimate error margin distinguish to 0.863%,-0.89%,-5.13%,-2.722%.The boiler that examines the result enunciation exploitation heredity calculate way and the BP network to combine exhausts the characteristic network model to refrain from rash action the speed,the study error margin compare of improvement of the model of the BP calculate way establishment all is the best.Through the medium of ANN erect large power station boiler NOx let characteristic model queen, namely approve as per feed prediction boiler let characteristic, in the event of incorporation global optimize algorithm, could scout out optima operating variable, with obtain low NOx discharge concentration out in for.