Optimizing a Cogeneration sCO₂ CSP-MED Plant Using Neural Networks

Defining optimal cogeneration system design requires the use of complex analyses capable of capturing dynamic processes within multiple subsystems and individual devices in parallel. This is due to the well-known fact that optimal cogeneration system performance does not always correlate with the optimal performance of a single subsystem. Furthermore, subsystems and single devices often present inherent design tradeoffs which are not easily captured in subsystem level models. Here, we perform a steady state thermodynamic and economic analysis for a concentrated solar power (CSP) cogeneration system producing power through a supercritical carbon dioxide (sCO₂) Brayton cycle and water through a multieffect distillation (MED) plant. The use of three artificial neural networks allows for the prediction of economic performance (levelized cost of water and electricity) and system performance (thermal efficiency, second law efficiency, performance ratio, and solar performance ratio). The cogeneration system results in a higher levelized cost of electricity (LCOE) than state-ofthe- art CSP sCO₂ plants. However, this reduction in power performance allows for a levelized cost of water of 1.1 $/m³, which is comparable to conventional membrane-based processes (1.25 $/m³) and significantly less than other solar thermal (1.8 $/m³) desalination systems. The nonparasitic integration between the sCO₂ Brayton cycle and MED plant also allows for maximization of water production without altering the second law efficiency of the cogeneration system, which remains at 11% during the analysis.