Robust model predictive control of condensate throttling in an ultra-supercritical plant for frequency supports of power systems with high renewable share

To enhance the load response capability of coal-fired power plants in transition to 100% renewable, a neural network (NN) based robust model predictive control (MPC) with accurate demand tracking of condensate throttling is developed. The immeasurable megawatt (MW) from throttling condensate which is filtered out from the total output power, is used to train the NN. A neural state observer (NSO) is then built to predict the immeasurable MW, with a robust stabilization technique applied to address the uncertainties introduced by the filtering process. Based on the NSO, a state-variable MPC is developed for condensate throttling, enabling robust load demand tracking while improving dynamic performance. The control design can be formulated as a nonlinear model-based H2/H∞ suboptimization problem. A solver utilizing linear matrix inequality (LMI) technology, is designed to solve this nonlinear complex problem without the need for linearization. To maximize the utilization of condensate throttling and minimize the required capacity of a battery energy storage system (BESS), a multi-scale load dispatching filter is developed under the virtual power plant (VPP) concept. The effectiveness of the approach is demonstrated through real-time simulations on the StarSim Hardware-in-the-Loop (HIL) Simulator.