Nonlinear model predictive control of organic Rankine cycles for automotive waste heat recovery: Is it worth the effort?

Abstract

Using organic Rankine cycles (ORC) for waste heat recovery in vehicles promises significant reductions in fuel consumption. Controlling the organic Rankine cycle, however, is difficult due to the highly transient exhaust gas conditions. To tackle this issue, nonlinear model predictive control (NMPC) has been proposed and approximate NMPC solutions have been investigated to reduce computational demand. Herein, we compare (i) an idealized economic NMPC (eNMPC) scheme as a benchmark to (ii) a NMPC enforcing minimal superheat and (iii) a PI controller with dynamic feed-forward term (PI-ff) in a control case study with highly transient disturbances. We show that, for an ORC system with supersonic turbine, the economic control problem can be reduced to a single-input single-output superheat tracking problem combined with a decoupled steady-state real-time optimization (RTO) of turbine operation, assuming an idealized condenser. Our results indicate that the NMPC enforcing minimal superheat provides good control performance with negligible losses in average power compared to the full solution of the economic NMPC problem and that even PI-ff only results in marginal losses in average power compared to the model-based controllers.