Multi-objective optimization of a SCO2 Brayton/zeotropic ORC combined cycle based on machine learning regression and classification

Research output: Contribution to journalArticlepeer-review

Abstract

This study introduces an innovative Supercritical CO2/Zeotropic ORC Combined System (SZOC) that substantially improves supercritical CO2 (SCO2) cycle performance through utilization of zeotropic working fluid temperature glide characteristics. The proposed system features an intermediate heat exchanger (IHX) as its central coupling mechanism, and integrates a recompression SCO2 Brayton cycle with a zeotropic organic Rankine cycle (ZORC) to achieve superior thermodynamic performance. The analysis methodology combines a thermodynamic modeling approach incorporating 1D radial turbine simulations with customized printed circuit heat exchanger (PCHE) models specifically designed for both SCO2 and ZORC applications. To recognize the system’s inherent nonlinear complexity, an advanced machine learning framework is implemented to develop a highly accurate surrogate model (R2 > 99.6 %) in conjunction with a specialized classification algorithm (97.8 % accuracy) for nonlinear constraint identification. The latter significantly enhances the optimization workflow. Through multi-objective optimization using the NSGA-II algorithm, the working fluid mass fraction ( Rm ) and split ratio ( Rs ), with thermal efficiency ( ηc ) and area-to-power ratio (APR) serving as primary performance metrics are simultaneously optimized. The present analysis reveals that R32/R245fa mixtures ( Rm ≈ 85 % R32) combined with a split ratio of Rs ≈ 0.2 demonstrate exceptional thermodynamic compatibility. The TOPSIS decision analysis has identified R32/R600a ( Rm = 88.2 %, Rs = 0.203) as the optimal configuration, delivering a remarkable combined cycle efficiency of 45.62 % − representing a 2.18 % absolute enhancement over conventional SCO2 systems (43.44 %), while increasing APR from 0.3965 to 1.3476 m/MW. These results demonstrate the advantages of zeotropic mixtures for advanced thermal system optimization and establish important foundations for next-generation SCO2 combined cycle development and provide actionable design guidelines for practical implementation.

Original languageEnglish
Article number128402
JournalApplied Thermal Engineering
Volume281
DOIs
StatePublished - 15 Dec 2025

Keywords

  • Genetic algorithms
  • Machine learning classification
  • Machine learning regression
  • SCO Brayton combined cycle
  • Zeotropic ORC

Fingerprint

Dive into the research topics of 'Multi-objective optimization of a SCO2 Brayton/zeotropic ORC combined cycle based on machine learning regression and classification'. Together they form a unique fingerprint.

Cite this