Nonlinear model predictive control of USC boiler-turbine power units in flexible operations via input convex neural network

Abstract

Coordinated control of an ultra-supercritical (USC) boiler-turbine unit in the flexible-operation mode is challenging because of internal nonlinearity, various unknown disturbances, and the strong-coupled multivariable behavior of the unit. A novel nonlinear model predictive control (NMPC) approach is proposed to improve the flexibility of the USC unit, which introduces an input convex neural network (ICNN) to model the dynamics of the USC unit, and then computes the optimal control decisions by solving a convex model predictive control problem. Simulations are conducted on a 1000 MW USC boiler-turbine unit. The results demonstrate that the ICNN exhibits equivalent performance to a conventional neural network in learning the system dynamics. The root mean square errors of the throttle steam pressure, the separator steam enthalpy, and the power are 0.003 MPa, 0.162 kJ/kg, and 0.513 MW, respectively. In load tracking and disturbance rejection simulations, the proposed ICNN-based NMPC outperforms the linear model predictive control and conventional neural network-based NMPC. The ICNN-based NMPC provides a fast and stable load tracking capacity at a ramp rate of 10.8% MCR/min. Moreover, the superiority of the proposed approach is further confirmed by comparison with the internal-model robust adaptive control method, which is a state-of-the-art method. The proposed approach has good potential to improve the operational flexibility of USC power units.