Improving Superheated Steam Temperature Control Using United Long Short Term Memory and MPC

Abstract

Superheated steam temperature is one of the most important process variables for controlling the steam quality of thermal power units. In order to improve the accuracy of superheated steam temperature and the stability of valves for desuperheating water, this paper proposed a novel control strategy called united long short-term memory (LSTM) and model predictive control (MPC), which is weighted by particle swarm optimization. First, a deeply learnt inverse model is made to express the potential nonlinear dynamic characteristics of data and to predict the future valve opening in short-term. Secondly, model predictive control is used to control the secondary superheated steam temperature. Thirdly, the two predicted valve opening are weighted by particle swarm optimization. The combined deep learning inverse model control and MPC can make up the deficiencies of each other, i.e., over fitting of deep learning inverse model and linearity of MPC. The simulation experiments proved the advantage of LSTM-MPC in comparison with traditional PID and single MPC control.