Lstm Modeling and Performance Analysis of Extended-State Kalman Filter-Based Energy-Saving Model Predictive Control for Supercritical Unit

Abstract

With the growing intermittency of renewable energy, supercritical coal-fired power plants (SCFPPs) are essential for power stability and efficiency. This study proposes an extended-state Kalman filter (ESKF)-based energy-saving model predictive control (EMPC) strategy to address the dynamic efficiency optimization challenges of SCFPPs. The long-short term memory model is constructed to capture the system dynamic characteristics, validated against the field data. The EMPC strategy minimizing tracking error and dynamic exergy loss is developed and integrated with ESKF for precise state estimation under disturbances and uncertainties. In rising load simulations, the ESKF-EMPC demonstrates superior performance in multi-loop tracking, with control index showing respective enhancements of 87.0%, 63.0%, and 62.9% over the conventional controller. Furthermore, the ESKF-EMPC reduces dynamic exergy loss by 63.1% and 12.8% compared to PID and MPC, respectively. Under disturbance cases, the ESKF-EMPC and ESKF-MPC strategies exhibit smaller control deviations compared to PID, and the ESKF-EMPC achieves a reduction in total exergy loss by 30.1%, 23.2%, and 7.2% compared to PID. Coal quality uncertainty simulation demonstrates that although the control performance of ESKF-EMPC is slightly inferior to ESKF-MPC, the average dynamic exergy loss of ESKF-EMPC is 1.8% lower than that of ESKF-MPC, which indicates its advantage in energy efficiency optimization.

Keywords: Long-short term memory (LSTM), Model predictive control (MPC), Extended-state Kalman filter (ESKF), Disturbance/Uncertainty rejection, Supercritical power plant