Thermodynamic design space data-mining and multi-objective optimization of SCO2 Brayton cycles

Highlights: • Quantitative effects of design parameters on Brayton Cycles' performances are drawn. • The SOM helps gain a deeper understanding of the design space of Brayton cycles. • A high-efficient hybrid algorithm is used to optimize Brayton cycles. • Data mining coupled with hybrid algorithm can boost the Brayton cycles' performances. This article implements the thermodynamic design space data-mining and multi-objective optimization of two typical supercritical carbon dioxide (SCO₂) Brayton cycles: the recompression Brayton cycle (SCO₂RBC) and the recompression reheating Brayton cycle (SCO₂RRBC). Firstly, a mathematical model with more constraints has been established for the two Brayton cycles. The maximum errors of the mathematical model relative to the references for the SCO₂RBC and SCO₂RRBC are 2.5%, 3.5% respectively. Then, three data-mining techniques (global sensitivity analysis by ANOVA, single factor analysis, coupling analysis by SOM) are successively applied to explore the design space. As a result, four key design parameters have been identified: the maximum and the minimum cycle temperatures, the pressure ratio, and the shunt flow percentage. And they present different non-linear effects on the cycles' performances (monotone increasing or decreasing, parabolic type with extreme point). It is also found that in order to achieve a global optimum, the maximum cycle temperature should be close to its upper bound, while the minimum cycle temperature tends to approach its lower bound, and a larger pressure ratio of compressor as well as a smaller shunt flow percentage is also required. Therefore, the data-mining methods are heuristic and can provide useful information for quickly searching the global optimums of SCO₂ Brayton Cycles. Finally, a hybrid optimization algorithm is introduced to optimize the Brayton cycles. It shows that the search efficiency of the hybrid algorithm is 3 ∼ 4 times higher than the traditional stochastic algorithms. For the given design space, the cycle efficiency of the SCO₂RRBC is improved by 10 percentage points. The hybrid algorithm coupled with the data-mining techniques are likely to speed up the design process of Brayton cycles, and have the potential to further improve the cycles' performances.