A review on machine learning techniques in thermodynamic cycle system design and control for energy harvesting

The supercritical CO₂ cycle and organic Rankine cycle are regarded as efficient energy conversion technologies. Current research is mainly focused on the working fluids, configurations, design parameters, components, dynamic performance, and control methods of thermodynamic cycle systems. A variety of parameter variables are involved in a complete optimization process, making the design of the optimal thermodynamic cycle systems become a highly complex problem. The machine learning technique has powerful predictive capabilities, and is expected to solve the problem with multiple variables. This paper provides a comprehensive review of machine learning methods applied in various design and operation levels of organic Rankine cycle and supercritical CO₂ cycle systems. Moreover, the approach to improving the interpretability of machine learning models is also reviewed, followed by the proposal of a system-wide holistic design framework for the thermodynamic cycle system. The framework views a complex global optimization problem as a mixed-integer nonlinear programming problem, where intelligent optimization algorithms and machine learning models assist in design. This study provides the first overview of all aspects of machine learning-based thermodynamic cycle system design and operation, which is of great significance for the intelligent design of such systems.