Skip to content

Applications

applications/ holds studies that run many TANGO simulations: design optimizations, surrogate-model pipelines and demonstrations. Each application starts from one configuration in templates/, varies it, and runs Workflow(cfg).run() once per design point, in parallel. Every application is a self-contained directory. Nothing outside the directory imports from it, and tango_core init --app <name> copies one into your own project.

Catalog

Application Kind Base configuration What it answers
demo_hydro_sf_electric Demo hydro_sf_electric.yml One run, plotted: well temperature profiles, reservoir temperature, power over time
demo_well_mesh_convergence Demo hydro_sf_electric.yml How fine the NumericalWell cell mesh must be; documents the default cell length
opt_* (one per template, 17 in all) Optimization The configuration of the same name Designs that minimize LCOE and maximize mean net power (electric, or thermal for direct heat); see Optimizations

On hold, under _on_hold/: opt_hydro_df_electric_surrogate, the opt_hydro_df_electric search on a trained surrogate model; see On hold.

Structure of an application

Kind Entry point Pipeline Also contains
Demo (demo_*) <name>.py none —
Optimization (opt_*) <name>.py dvc.yaml params.yaml, SLURM <name>_job.sh

Results go to the application's outputs/ directory, or to --save-dir where the script supports it. DVC stages depend on src/tango_core.hash, so a change to the package invalidates them automatically.

Running

Smoke-test every application end to end (tiny designs, temporary directories, nothing written to outputs/):

just apps

Run one application:

# Demo
uv run python applications/demo_hydro_sf_electric/demo_hydro_sf_electric.py

# Optimization, tiny check into a throwaway directory
uv run python applications/opt_hydro_sf_electric/opt_hydro_sf_electric.py \
    --smoke --save-dir /tmp/opt-check --no-plot

On the ETH Euler cluster, submit full-size runs with just slurm <app> ... from anywhere in the checkout. See Running on Euler.

Flag Applies to Effect
--smoke optimizations A 2-generation, population-6 search with any constraint relaxed
--clean optimizations Delete generated outputs before running
--n-processes optimizations Parallel worker count
--n-max-gen, --pop-size optimizations NSGA-II generations and population size
--no-plot, --plot-only optimizations Split the search from the plotting, as the DVC stages do
--save-dir optimizations Where results and checkpoints go
--skip-param-check optimizations Skip the check that every decision variable matters

Optimizations

Each opt_* script defines its decision variables, bounds and objectives and hands them to tango_core.pymoo_api.OptApp, which supplies the command line, the baseline run, the NSGA-II search (tango_core.pymoo_api.Minimization) and the plots. Each has its own DVC pipeline and SLURM job script.

Every application searches mass flow rate, injection pressure and the surface pipe diameter. Doublets also search the injection and production well diameters, plus the target extraction pressure where the production well has a pump. AGS applications also search well depth, horizontal loop length and diameter, and vertical well diameter. Condensing pressures for water and ORC cycles are site conditions and are not searched. Each application also reports production temperature, specific investment cost and OPEX for its Pareto front.

Application Plant-specific variables
opt_hydro_sf_electric, opt_hydro_sf_electric_5spot, opt_EGS_electric Flash (separator) pressure
opt_hydro_df_electric High-pressure flash pressure, low-pressure flash as a fraction of it
opt_hydro_ORC_electric, opt_AGS_ORC_electric, opt_CO2_EGS_ORC_electric ORC feed-pump (evaporating) pressure
opt_CO2_AGS_ORC_electric ORC feed-pump pressure, preheater CO2 outlet pressure
opt_CO2_EGS_electric, opt_CPG_electric, opt_CO2_AGS_electric CO2 condensing (turbine exhaust) pressure
opt_CO2_AGS_direct_heat District-heating exchanger outlet pressure
opt_hydro_direct_heat, opt_EGS_direct_heat, opt_CO2_EGS_direct_heat, opt_CPG_direct_heat None
opt_AGS_direct_heat None; also constrains production temperature to at least 105 °C

Before searching, every application checks that each decision variable matters: moving it up and down from the configuration's own value must change at least one objective or constraint, or the run stops with InertParameterError naming it. This costs 1 + 2 × (number of variables) extra simulations; skip it with --skip-param-check once an application is known to be good.

On hold

Applications in _on_hold/ are kept but no longer run: just slurm, just apps, the wheel and tango_core init all skip them.

  • opt_hydro_df_electric_surrogate: opt_hydro_df_electric's search against a trained surrogate model instead of the real simulation. Its LCOE predictions near the Pareto front never got within the 1% needed to trust the result. Whether to bring it back or delete it is tracked in #106.

Adding an application

  1. Copy the closest existing application directory and rename every file prefix to the new name.
  2. Point it at a configuration in templates/. Add a new configuration there first if none fits.
  3. Give it a fast end-to-end mode with --smoke. The apps recipe in the justfile finds every opt_* application and runs it with --smoke; add any other application's invocation there by hand.
  4. List every file in pyproject.toml's [tool.hatch.build.targets.wheel.force-include], so that tango_core init --app ships it.
  5. Add a row to the catalog above.