Data-driven nonlinear modeling for superheating degree in organic Rankine cycle systems

Organic Rankine Cycle (ORC) power systems have become a promising solution for improving energy
efficiency, particularly in waste heat recovery (WHR) applications. These systems convert low-grade
heat into electricity, contributing to energy savings and emission reductions. A key control objective is
the regulation of the superheating degree, as it directly affects thermodynamic performance and system
reliability. This work explores a data-driven modeling strategy using neural networks (NNs) to capture
the nonlinear dynamics of the superheating process in a small-scale (11 kWel) ORC unit. To enhance
generalization and interpretability, an automatic feature selection framework based on reinforcement
learning is developed. The approach evaluates the relevance of multiple candidate input variables, selecting
the most informative ones to optimize predictive accuracy. Experimental results show that the
proposed model effectively reproduces system behavior while maintaining strong generalization to unseen
operating conditions. This modeling framework lays the foundation for advanced control development
and contributes to data-driven methodologies for energy systems optimization