Abstract: As one of the major environmental protection indicators, the value of NO_x emission from fossil fuel power plants has been supervised and assessed strictly in real time by the national environment protection department. Unfortunately, due to the intrinsic long duration of pure time delay dynamics in the NO_x controlled model, regulating the NO_x emissions within the ideal range by conventional PID controls would not be easy. In this paper, a strategy of the predictive dynamic matrix control (DMC) algorithm combined with BP neural network model is proposed for the denitration control, in which the BP neural network is implemented to approach the response of the zero predictive input in the DMC control. Therefore, by taking advantage of the BP generalization ability, the time-variant characteristics of the model at different loads in real industrial processes can be approximated to make the predictive model precisely closer to the practical object and consequently improve the precision of the predictive model output in the DMC algorithm. The application in a fossil fuel power plant shows that the new control strategy can effectively improve the NO_x control performance.

Key words: fossil fuel power plant, denitration control, neural network, predictive control