An LSTM-PSO-Based Predictive Strategy and Its Application in the Main Steam Temperature Control in Power Plants

Abstract

Abstract We proposed an LSTM-PSO predictive controller strategy to overcome the large delay and strong disturbance during main steam temperature control. First, the long short-term memory (LSTM) network model was developed to forecast the future trend of mainstream temperature, and the deviation between the model output and the expected value after feedback correction was minimized by the particle swarm optimization (PSO) algorithm. Then, the LSTM-PSO predictive controller was taken as the primary controller to maintain the steam temperature of the coal-fired boiler. A comparative simulation with conventional PID cascade control and dynamic matrix controller strategies was given, and the test calculations indicate that the adjustment time was shortened and the overshoot was reduced. Finally, the LSTM – PSO predictive control strategy was deployed into the actual boiler operation process of a 300 MW coal-fired power plant, within 0.5 °C steady-state error and 2 °C dynamic error.