Configuration Guide#

One run of workflow/main.py solves one scenario with the four storage formulations. Everything it needs is read from config/config.yaml; all paths in that file are relative to the repository root.

The reproduction tools (notebooks/reproduce_study.ipynb, workflow/automation.py) rewrite the scenario settings for each run and restore the file afterwards, so it only needs editing by hand for single runs.

Scenario#

scenario_name: k4h1WND        # label used in every output file name
resource: WND                 # WND | SOL | HYD (labels and the k365h1 baseline file)
n_clusters: 4                 # representative days k (365 = full-year benchmark)
hour_grouping: 1              # block length h in hours
days_in_year: 365
seasons: 1
random_state: 0               # k-means seed
vre_technology: WINDPOWER     # technology label in the capacity-factor files

Key	Meaning	Values used in the study
`scenario_name`	Prefix of all result files. Convention: `k<k>h<h><resource>`, suffix `_limited` for the stress test, `_s<seed>` for other seeds	`k4h1WND`, `k20h1SOL`, `k365h1HYD_limited`, …
`resource`	Which system the run represents; also selects the matching `k365h1` baseline for the figures	`WND`, `SOL`, `HYD`
`n_clusters`	Number of representative days k	2–10, 20–100 in steps of 10, and 365
`hour_grouping`	Hours per intra-day block h; must divide 24	1, 2, 4, 6, 8, 12
`days_in_year`	Length of the chronological year in days	365 (do not change: the time series have 8760 hours)
`seasons`	Number of seasons in the Welsch formulation	1
`random_state`	k-means seed (`n_init = 1`, k-means++)	0; 1 and 2 for the seed-sensitivity runs
`vre_technology`	Technology whose new capacity is reported as `Power Capacity`	`WINDPOWER` (all three profiles use this label)

Derived quantities: blocks per day b = 24 / h, representative time slices k · b, chronological time blocks 365 · b.

Storage#

StorageMaxCapacity: 1000.0     # cap on total storage energy capacity, kW·yr
ResidualStorageCapacity: 0.05  # existing storage, kW·yr
StorageLevelStart: 0.05        # not applied (the initial SoC is a decision variable)

StorageMaxCapacity caps total capacity (new + residual). 1000 kW·yr is non-binding and is used for all main runs; 0.2 kW·yr gives the binding-capacity stress test (reported new capacity 0.15).

Inputs#

system_inputs: data/system      # OSeMOSYS parameter CSVs shared by all formulations
data_8760:
  directory: data/timeseries
  CF: CapacityFactor_Wind.csv   # or CapacityFactor_Solar.csv / CapacityFactor_Hydro.csv
  SDP: SpecifiedDemandProfile.csv

CF and SDP are the 8760-hour profiles that are clustered and aggregated. To study another system, add the new profiles to data/timeseries/ (same CSV layout) and point CF/SDP at them.

Outputs#

results:
  directory: results                    # all result files of the run
  directory_GUI: results/GUI            # copy of the hourly comparison figure
  results_copied_filename: Storage_Level
  results_excel_file: Simulation_Results.xlsx

Repeated runs append rows to Simulation_Results.xlsx. The files are described in Outputs.

Formulations#

The cases block defines one entry per formulation (Model_Cluster is Novo): its folder under models/, model file, otoole template and the names of the generated CSVs. These entries match the model files and normally need no change.

To run only some formulations, add a list:

models: [Model_Niet, Model_Kotzur]

Without it all four run. The comparison figures are produced only when all four run.