Model Specifications#

Test system#

All runs use the same single-node OSeMOSYS system, defined in data/system/:

Element	Value
Region, year	`Simplicity2`, one model year (2010)
Fuel	`ELECTRICITY`, annual demand 20, shaped by the hourly profile in `data/timeseries/SpecifiedDemandProfile.csv`
VRE technology	`WINDPOWER` (the label is the same for all three systems; only the capacity-factor profile changes: wind, solar or hydro)
Storage	`BATTERY`, charged and discharged by the technology `BATTERY_STORAGE` (mode 1 charges, mode 2 discharges)
Storage cost and life	`CapitalCostStorage` 1000 per kW·yr, `OperationalLifeStorage` 1 year, `MinStorageCharge` 0 (otoole defaults)
Residual storage	0.05 kW·yr (`ResidualStorageCapacity` in `config/config.yaml`)
Storage energy cap	1000 kW·yr (non-binding) in the main runs; 0.2 kW·yr in the stress test

The initial state of charge is a decision variable: StorageLevelStart in config/config.yaml is not applied. All values are listed with their source in step 2 of the reproduction notebook.

Formulations and model files#

Each formulation has its own folder in models/ with an OSeMOSYS model file and an otoole configuration template. They share the rest of OSeMOSYS and differ only in the time sets and the storage equations.

Formulation	Folder	Extra time sets / mappings	SoC equation	Bounds
Niet	`models/Model_Niet`	none beyond `TIMESLICE`	`S2_StorageLevelTSStart` over representative time slices, weighted by `YearSplit`	`SC1`, `SC2` at each time slice
Welsch	`models/Model_Welsch`	`SEASON`, `DAYTYPE`, `DAILYTIMEBRACKET`, `DaysInDayType`, `Conversionls/ld/lh`	`S5`–`S15` (season, day-type start and finish)	`SC1`–`SC4` at the first and last day-type instances
Kotzur	`models/Model_Kotzur`	`DAYSCRO` (365 chronological days), `Conversionldc`	`S2_StorageLevelDayTypeFinish`, `S3_StorageLevelChronoDayStart`	`SC1`, `SC2` at each chronological day
Novo	`models/Model_Cluster`	`TIMESLICECRO` (365 × 24/h blocks), `Conversionlts`	`S2_StorageLevelTSStart` over chronological blocks	`SC1`, `SC2` at each chronological block

All four require the storage to end the year where it started (a refilling constraint: SC8 in Niet and Novo, SC4 in Kotzur, S15 in Welsch) and cap the total storage capacity at StorageMaxCapacity (SC3, or SC7 in Welsch). See About Storage Modelling for what these differences mean in practice.

How the inputs are generated#

workflow/main.py writes the time sets and mappings each formulation needs into models/Model_<name>/inputs_csv/ (generated, not tracked), copies the shared parameters from data/system/, and converts everything to a GLPK data file with otoole. Run-specific values (DaySplit, YearSplit, StorageMaxCapacity, ResidualStorageCapacity) go into a temporary otoole_config_<name>.run.yaml, so the templates in models/ never change. The steps are described in Workflow Guide.

Solver#

GLPK 5.0 (glpsol). If GLPK reports a non-optimal result, the run is retried with --nopresol, --dual and --exact; failed solves are recorded with status FAILED and no capacities.