Multi-objective performance optimization of regenerative S-CO2 Brayton cycle based on neural network prediction

The regenerative supercritical CO2 Brayton cycle (RSCBC) has great development potential in waste heat recovery and utilization, and it is necessary to carry out performance analysis and optimization. This article first applies the theory of finite time thermodynamics to establish a RSCBC model with finite temperature difference heat transfer, irreversible compression, irreversible expansion and other irreversible factors under variable temperature heat source conditions, and then uses the data samples to construct the corresponding neural network model. Based on the NSGA-II algorithm, the working fluid mass flow rate, the pressure ratio, the heat conductance distribution ratios of the regenerator and the heater are chosen as optimization variables, multiobjective optimization is carried out with the goals of cycle thermal efficiency, net power output, ecological function and exergy efficiency. The results show that the use of neural network models to predict cycle performance can save a lot of calculation time compared to traditional calculation methods; after optimization, the positive ideal point is not on the Pareto front, which shows that the four optimization objectives are mutually restricted and affect each other. The results by using Shannon Entropy method for decision-making have a lower deviation index, and those by using TOPSIS and LINMAP methods for decision-making are consistent with each other; for the results by using Shannon Entropy method for decision-making, the cycle thermal efficiency and net power output can reach 38.4% and 12.047 MW respectively, which can be increased by 31.15% and 43.29% compared to those for the initial design point respectively, and the ecological function and exergy efficiency can reach 7.6274 MW and 73.2% respectively, which are 4.2588 times and 33.75% higher than those for the initial design point respectively. The obtained results can provide some guidance for the optimal design of the RSCBC in real engineering.