Neural network based superheater steam temperature control for a large-scale supercritical boiler unit

Abstract

To improve the Superheater Steam Temperature (SST) control of a large-scale supercritical coal-fired boiler generating unit, this paper presents an inverse compensation control scheme based on dynamic recurrent neural network (NN) inverse process models for the multi-stage water-spray desuperheating controllers. With proper analysis of the boiler design and operational characteristics, the inputs and outputs of the NN inverse models are determined. Then two NN inverse dynamic models of the superheater system are trained and validated with historical operating data over a wide-range of loading conditions. The trained NN inverse models are then employed as internal model controllers to improve the SST control effect by providing real-time supplementary signals to the original cascade PID controllers. The real-time steam temperature signals are fed back to adjust the input reference values of the NN controllers automatically. The controller is programmed in MATLAB and communicates with a full-scope simulator of a 600MW supercritical coal-fired power generating unit. Detailed simulation tests are carried out, which shows the new compensation control scheme can dramatically improve the SST control of the supercritical boiler.