Towards model predictive control of supercritical CO2 cycles

TL;DR

This paper develops a model predictive control approach for supercritical CO2 power cycles, effectively managing complex system interactions and constraints, and demonstrating high performance in simulations.

Contribution

It introduces a control methodology combining high-fidelity modeling and online linearization for supercritical CO2 cycles, enabling effective MPC implementation without parameter scheduling.

Findings

Controller maintains high turbine inlet temperatures during load changes

Achieves net power output ramp rates over 100% per minute

Performs well across 35-105% of capacity without parameter tuning

Abstract

Control of non-condensing non-ideal-gas power cycles is challenging because their output power dynamics depend on complex system interactions, non-ideal-gas effects complicate turbomachinery behavior, and state constraints must be respected. This article presents a control methodology for these systems, comprising a control modeling approach and model predictive control (MPC) strategy. This methodology is demonstrated on the high-pressure side of a simple supercritical CO2 cycle power block, composed of a variable-speed compressor, heat exchanger, and fixed-speed turbine. The control model is developed by applying timescale-separation arguments to a high-fidelity simulation model and locally linearizing non-ideal-gas turbomachinery performance maps. MPC is implemented by linearizing the control model online at each sampling instant. Closed-loop simulations with a full-order gas-dynamics truth model demonstrate the effectiveness of the proposed control methodology. In response to load changes, the controller maintains high turbine inlet temperatures while achieving net power output ramp rates in excess of 100% of nameplate output per minute. The controller often acts at the intersection of motor torque, compressor surge, and turbine inlet temperature constraints, and performs well from 35 to 105% of nameplate capacity with no parameter scheduling. Good performance and fast update rates are obtained via online linearization. The results demonstrate the suitability of MPC for the supercritical CO2 cycle, and the proposed methodology is extensible to more complex cycle variants such as the recuperated and recompression cycle.