Efficient techno-economic optimization of integrated energy systems using Bayesian optimization

Energy infrastructure needs are being met with the development of various technologies including thermal energy storage, modular nuclear power, and steam electrolysis. Economic optimization of these technologies, or techno-economic optimization, is currently done with a black box gradient descent approach. The expense of gradient estimation and local nature of the algorithm motivate developing, implementing, and analyzing a suitable alternative. To address this, Bayesian optimization is implemented within Idaho National Laboratory’s RAVEN and HERON framework and applied to the Natrium energy system. The best and worst Bayesian optimization variations are shown to outperform gradient descent in evaluation efficiency, solution quality, and robustness to initial conditions; therefore, Bayesian optimization is an improvement over gradient descent in all scenarios tested. Within the California (CAISO) market, the Natrium system generates an average net annual cashflow of $13 million with a plant capacity of 0.426 GWe and thermal storage capacity of 3.22 GWth-hours. Within the mid-continent (MISO) and Texas (ERCOT) markets, the Natrium system is not economically viable generating average net annual cashflows of -$47 million and -$35 million, respectively.