Graph-based configuration optimization for S-CO₂ power generation systems

Configuration optimization for energy conversion systems is still under-researched compared with that in the chemical process area. As the basis of design and operation optimization, configuration optimization helps generate more advanced thermodynamic configurations and dramatically promotes the efficiency of thermal systems. However, the variable in configuration optimization involves discrete variables to make optimization more complicate when mixed with continuous design variables. Current configuration optimization research focuses on task-orient methods, superstructure, and heuristic algorithms. These methods tend to screen out candidates based on high levels of knowledge from researchers. In this paper, we first developed a graph-based configuration representation with minimum thermodynamic knowledge requirements. Therefore, the searching space shrinks without losing any possible optimum point. The representation is embedded in a simulated annealing-based optimizer. This optimization framework was validated by optimizing the S-CO₂ power generation system's configurations with simple and complex component number limitations. This framework has an over 99% generation rate of feasible configurations. The simple and complex case study results show that the optimization framework converges to a recuperated system with an efficiency of 38% and a recompression system with 44% under maximum and minimum temperature of 850 K and 310 K. Moreover, the result of the simple case study was further justified with an exhaust search. Moreover, both case studies' convergence stabilities are confirmed by multiple and independent framework executions. Finally, the initial configuration's influence shows that the best configuration on lower volume searching space will significantly help with computational efficiency and whole converge process on high volume search space problem.