|
Nitrogen oxide (NOx) emissions from coal-fired utility boilers major pollutants. With the improvement of environmental requirements, the modern power companies face lower operating costs and reduce emissions dual requirements. Boiler is a complex multi-variable system, the NOx emission characteristics of complexity, it is difficult to estimate with a simple formula, often based on test results fumble NOx reduction method. But the real scene furnace test workload, limited test conditions, the parameters have influence on NOx emissions, and superimposed on each other, leading to difficulties in data analysis, and experimental results can not be obtained under estimation formula and specific calculation model, the test results can not be further promotion. Neural network modeling is an important feature is its black box between input and output characteristics, if the boiler as a black box, it must determine the input must correspond to the output, so the application of artificial neural network modeling of emission characteristics of the boiler. Based on the analysis of the coal-fired boiler NOx formation and destruction mechanism, based on the discussion of the impact of coal-fired boiler NOx emission factors. Using the boiler thermal test data, using three-layer BP neural network model was constructed boiler emission characteristics. BP learning algorithm for conventional low efficiency, slow convergence defects, batch learning, additional momentum, adaptive learning coefficient and other measures to improve. The measured data validation through the boiler, BP neural network NOx, exhaust gas temperature, the carbon content of fly ash, smoke oxygen relative prediction error of 0.49% -1.954%, 8.14% -5.134%. From the test results, NOx, exhaust gas temperature predictions are very close to the measured values. For the initial weights and thresholds of the network convergence speed and error precision greater impact on this issue, using real-coded genetic algorithm for network weights optimization first, then use the improved BP neural network optimization. Utilization of genetic algorithm global search ability and local search ability BP neural network characteristics. Designed and implemented based on genetic algorithm and BP network combined with the boiler emission characteristics of the network model, GA-BP network for NOx, exhaust gas temperature, fly ash carbon content, exhaust oxygen content relative prediction error of 0.863%, -0.89 % -5.13% -2.722%. Test results show that the use of combined genetic algorithm and BP neural network model of boiler emission characteristics of network convergence speed, learning error than the improved BP algorithm are best established model. Through artificial neural networks to build large power plant boilers NOx emission model, based on the input parameters can predict emission characteristics of the boiler, if the combination of global optimization algorithm, you can find out the optimum operating parameters to achieve low NOx emission levels.
|