Investigation Of Operation Strategy On Low NOx Combustion Optimization Of Coal Fired Utility Boiler

Low NOx combustion optimization of boiler is to improve combustion efficiency and reduce NOx emissions through rational combustion in the furnace. But controlling measurements of NOx formation is opposite to steady combustion and burnout of fuel in the furnace. So how to give attention to NOx emissions and boiler efficiency and make cost of draining contamination and coal consumption lowest is the objective of combustion. Techinically, advanced control methods of higher efficiency and lower NOx emissions can be considered as consisiting of two stages. In the first stage some form of plant modeling, which uses both historical and current plant data, attempts to capture the relationship between the plants manipulated input variables and the NOx output. In the second stage some form of constrained optimization is used to manipulate the inputs of the model in order to minimize the NOx output. These values are then presented to the operator (open-loop mode) and guiding operation or in some cases used to automatically adjust the inputs (closed-loop mode).In this paper, studies of low NOx combustion optimization are firstly discussed and summarized. It mainly contains mechanisms of NOx formation and some typical control methods, and some dominating operation factors influencing NOx formation, and prediction methods of NOx emissions, and some optimizing algorithms.This paper develops experience prediction models of NOx emissions and boiler efficiency versus some input manipulated operation variables by bp neural network with an improved BP algorithms using L-M learning function based bayes regulation. The research objectives of modeling are two 300MW tangentially coal-fired boilers. And training data are results of combustion modification experiments on the two boilers. After the two models are successfully developed, testing experiments about predicting ability of model are proceeding. Testing results show that the model could quickly and accurately predict NOx emissions and boiler efficiency of real-time operating conditions.BP neural network have been widely applied to modeling and controlling of nonlinear system. And it has also been applied to controlling NOx emissions of coal-fired utility boilers. But before a successfully trained prediction model is got, it requires supplying abundant and information-rich historical data and premeditatedly adds some real-time operation data to the model. So that the model could be constantly updated to fully reflect dynamic operation conditions of boiler. But BP learning algorithm is a gradient descent algorithm which generally has problems of time-consuming and over fitting during neural network training. Thus model's abilities of updating and generalization are limited. To solve the problem, the paper developed prediction models of NOx emissions and boiler efficiency by least square support vector machine, which has advantages of quicker computation speed and better generalization performance. And a comparison about predicting ability was made between least square support vector machine model and BP model. Results showed that former models are able to accurately predict NOx emissions under different operation conditions and have a better generalization performance. Compared to other modeling methods, least square support vector machine is more suitable for on-line work.After successfully modeling NOx emissions and boiler efficiency, the paper combines the developed models with genetic algorithms and then achieves optimization searching of manipulated input operation variables fewer than three different optimization objectives. The three objectives are separately getting the lowest NOx emissions with unconcern for boiler efficiency, getting the highest boiler efficiency with unconcern for NOx emissions and giving attention to both NOx emissions and boiler efficiency. The optimal operation programme got from optimizing system has a practical feasibility. The whole optimizing process can supply model basis for closed-loop or open-loop control of low NOx emissions combustion optimization operation.