Intelligent modified predictive optimal control of reheater steam temperature in a large-scale boiler unit

A modified predictive optimal control (MPOC) scheme based on neural network modeling and particle swarm optimization (PSO) techniques is proposed in this paper for reheater steam temperature (RST) control of a large-scale boiler unit. A recurrent neural network is trained to directly model the temperature dynamic response of the reheater system. The neural network direct model is then used to evaluate the performance of the MPOC in search of the optimal control, where optimization is carried out with the PSO. A simplified PSO algorithm with search direction control is designed to find the nearest and optimal controls for the reheater steam temperature. To further improve the optimal search accuracy, each last-step prediction error between the direct model output and the actual RST is added to the current-step cost function to compensate for the model error. Control tests on a full-scope simulator of a large scale power generating unit have shown the validity of the proposed method.