An effective multi-model based nonlinear control for USC power plant

Energy efficiency optimization for the ultra supercritical (USC) boiler-turbine unit is a major concern in the field of power generation. In order to deal with the nonlinearity and slow dynamic response problems, a new nonlinear control method is proposed which integrates internal model control (IMC) and generalized predictive control (GPC) into a unified framework. Specifically, through a long short-term memory (LSTM) neural network based IMC, the system achieves rapid convergence to the vicinity of the desired setpoint, significantly enhancing the response speed. Then, by a composite weighted human learning optimization network based nonlinear generalized predictive control (CWHLO-GPC), high-accuracy tracking performance is achieved. Finally, an example on a 1000MW USC power plant demonstrates the proposed method can achieve fast and stable dynamic response under large load variation.