Supplementary data for: "Comparing generative process synthesis approaches with superstructure optimization for the conception of supercritical CO2 Brayton cycles"
Description
Supplementary data for the publication "Comparing generative process synthesis approaches with superstructure optimization for the conception of supercritical CO2 Brayton cycles" (by Antonio Rocha Azevedo, Tahar Nabil, Valentin Loubière, Romain Privat, Thibaut Neveux and Jean-Marc Commenge), currently under peer review. The pre-print of the paper may be found at: https://www.ssrn.com/abstract=5200900 This includes: - A dataset of > 400.000 randomly generated power cycles (Random_SFILES_all.csv), in both the SFILES 2.0 and modified SFILES notations discussed in the article; - Results of the generative process synthesis approaches (Evolutionary Programming and Machine Learning). It contains data for all generated cycles which have a positive cycle efficiency and an LCOE under 200$/MWh (= 3.62 in the normalized values presented in the publication)). Datasets are named according to experiment name and objective function optimized. All datasets include, for each process: • Efficiency and LCOE values (attention to the objective function being optimized); • The total number of simulations run during optimization; • Real time elapsed during optimization (hours); • CPU time elapsed during optimization (hours); • [Machine Learning only] Modified SFILES generated by the model; • SFILES 2.0 and modified SFILES representations for the optimized process; • A "parameter SFILES 2.0" representation of the process, which embeds unit operation parameters into the SFILES 2.0 string (modified SFILES representations are not available as we may not guarantee that its syntax is completely respected). For more information and examples, check the README.txt file; • SFILES 2.0 and modified SFILES representation of the post-treated process (after bypassed units and branches are removed, according to the procedure described in the manuscript's supplementary material); • [Evolutionary Programming only] The generation in which the process is generated; • [Machine Learning only] The iteration in which the process is generated. - UnitsData.json, which includes the unit operation parameters involved in simulation and optimization, as well as their default values and bounds; - A README.txt with examples on how to read UnitsData.json and parameter SFILES. Notes: - Post-treated SFILES may eventually contain errors/not accurately represent the underlying process, as the post-treatment procedure is done automatically. - When conversion to modified SFILES was not possible, the associated entry from the SFILES 2.0 column is used. - In case of errors with the generation of the SFILES representation (e.g., for evolutionary approach results or during post-treatment), the "(null)(out)" SFILES placeholder isused (represents an empty process).