Data-Driven model identification and efficient MPC via quasi-linear parameter varying representation for ORC waste heat recovery system

Organic Rankine Cycle (ORC) stands out in low-grade waste heat recovery (WHR) technology for its significant performance. Considering the system’s coupling dynamics, model predictive control (MPC) has adopted its wide application in realizing reasonable adjustment of the ORC based WHR system and has been proved to be effective. MPC is usually applied under the premise of having established a relatively accurate model, thus achieving a satisfactory control performance. However, the popular first principles model of the ORC system turns out to be high-dimensional and will result in computationally costly during typical nonlinear MPC adoption. While model linearization around operating points would enable the employment of linear MPC and reduce online calculation amount to some extent, the obtained local model loses global validity, leading to possible unstable control performance. To address these problems, a practical input–output data-driven quasi-linear parameter varying (QLPV) model is constructed for the ORC based WHR system by introducing the Koopman operator to ensure the global control effect. The corresponding MPC algorithm is thus presented via QLPV representation which solves the constructed constrained optimization problem iteratively in the form of a series of quadratic programming (QP) problems at each time step. Moreover, considering the possible lack of adequate training data covering the important dynamics of the ORC based WHR systems in practical application, an online updating mechanism that involves recursive equations is proposed to realize prediction accuracy improvement. Simulations on prediction, setpoint tracking and disturbance rejection are performed to verify the established model accuracy and the control effectiveness of the proposed strategy.