MultiObjectiveWorkflow

Work-in-progress tools for multi-objective optimization of Drogon-style energy system cases (wind + gas and wind + hydrogen scenarios).

Install

From the project root:

python -m pip install -e .

Optional extras:

python -m pip install -e .[analysis]
python -m pip install -e .[dev]

Note: runtime dependencies used by optimization backends (for example PET-related packages such as popt/subsurface) are expected to be available in your environment.

Examples

Both examples run a Pareto sweep over the weights in objective_weights: for each weight w the optimizer minimizes w * f1 + (1 - w) * (-NPV), where f1 is total CO2 emissions or CAPEX. Results for each weight go to <save_folder>/weight_<w>/, and plots are written after the sweep.

Run the examples from the examples directory (the scripts use relative paths):

cd examples
python run-wind_gas.py
python run-wind_hydrogen.py

User-editable variables at the top of each script:

• energy_sources: energy sources in the case (checked against the script's case)
• random_seed: random seed for reproducibility
• objective_weights: weights to run for the Pareto sweep
• save_folder: where results are written
• parallel: number of simulations to run in parallel
• run_optimization: set to False to skip optimization and only plot existing results
• show_plots: show figures on screen after plotting (figures are saved either way)

Wind + gas (run-wind_gas.py)

Offshore power comes from wind turbines, with gas turbines covering any shortfall. The controls are the time-varying rates of the seven Drogon wells (RATE_A1–RATE_A6, RATE_OP5). The two objectives are NPV and total CO2 emissions from the gas turbines. Emissions are costed in the NPV, and fuel gas is subtracted from exported gas. Config: data/config/wind_gas.py. Default output: outputs/wind_gas/.

Wind + hydrogen (run-wind_hydrogen.py)

Wind power, with hydrogen storage (electrolysis and fuel cell) covering shortfalls instead of gas turbines. On top of the well rates, it optimizes the number of wind turbines (N_WT) and the hydrogen storage capacity (N_H2). The two objectives are NPV and CAPEX, and a penalty (EPF) handles hydrogen shortfalls. Config: data/config/wind_hydrogen.py. Default output: outputs/wind_hydrogen/.

Plots

Plotting code lives in examples/plot_results.py. For each weight, figures are written to weight_<w>/figures/:

• obj_func.png (or obj_func_epf.png with penalty loops): objective history
• variable_<i>.png: initial vs. final well-rate controls
• free_variable_<i>.png (hydrogen only): initial vs. final N_WT / N_H2
• pareto_point.png:
  • Gas: histograms of NPV and total CO2 across the ensemble
  • Hydrogen: NPV histogram and CAPEX breakdown

In save_folder itself:

• pareto_curve.png: mean NPV against CO2 (gas) or CAPEX (hydrogen), one point per weight
• wind_power_profiles.png: sample wind power profiles

Closed-loop workflow

The closed_loop package holds the closed-loop short-term optimization (ECMOR 2026). It needs OPM Flow and the closed-loop extra:

python -m pip install -e .[closed-loop]
cd examples
python run-target_optimization.py   # optional: recompute the target rates (needs PET)
python run-closed_loop.py

• closed_loop.target_optimization: life-cycle (NPV) optimization of the target rates with PET (run.py), and target_controls.py, which simulates the optimum and writes its monthly rates. examples/run-target_optimization.py runs both in outputs/target_optimization/. Copy the resulting target_controls.csv to data/closed_loop/ to use it in the loop.
• closed_loop.workflow: the closed loop with the RBF proxy, run by examples/run-closed_loop.py. See src/closed_loop/workflow/README.md.

Inputs:

• data/config/target_optimization.yaml, data/config/closed_loop.yaml: settings.
• data/closed_loop/: the two deck templates, the initial and target rates, and the wind power.
• data/drogon_ensemble/Drogon_Mean/: the mean Drogon model, with the restart files of its history run that start the loop. The decks read it and include/ from data/drogon_ensemble/.

Tests

python -m pip install -e .[dev,analysis]
python -m pytest

The tests cover src/facility, src/multi_objective_optimization and examples/. They use small synthetic data and stand-ins for PET and OPM Flow, so no reservoir simulations are run.