NAGA — Network for ASEAN Grid Analysis

NAGA (Network for ASEAN Grid Analysis) is a standalone, anyone-can-run capacity-expansion and dispatch model of the ASEAN Power Grid (APG) at sub-national zonal resolution, solved by Benders decomposition with a monolithic reference solver for verification. Julia / JuMP, runs on the open-source HiGHS solver out of the box (Gurobi optional).

⚠️ Preliminary dataset — not for citation. The bundled data_asean/ uses a preliminary fleet derived from Global Energy Monitor (CC BY 4.0) and proxy/literature sources for several layers; an in-house dataset is in preparation and will replace it. Treat current numbers as illustrative. See DATA_SOURCES.md for provenance and attribution. The model code and method are stable and verified.

The model quantifies the value of regional grid coordination — and how that value changes when decarbonization pressure is externally imposed (EU CBAM carbon border pricing, corporate 24/7 carbon-free-energy procurement) rather than set by domestic targets. External pressure is an optional module, off by default — out of the box NAGA runs as a conventional regional coordination and planning model; enable the module to study CBAM and 24/7 CFE.

• Grid planners (AIMS III / utilities): a coordination ladder (Bilateral → MarketIntegration → PartialCoordination → FullCoordination) isolates the value of each mechanism — trade friction, shared adequacy, joint transmission and generation planning — and every run exports corridor shadow values ($/MW-yr of interconnection capacity), a demand-side corridor-prioritization criterion.
• MDBs / financiers: per-country cost, capacity, carbon-intensity and CBAM-exposure breakdowns show how coordination benefits are distributed, and adequacy shadow prices put a number on interconnection's reserve value.
• Corporates / 24/7 CFE buyers: hourly clean-energy matching (three modes, including the Riepin & Brown 2024 CFE score) with the import-eligibility question as a switch — whether cross-border clean energy counts toward CFE.

You need Julia (≥ 1.9) and Python 3. No commercial solver or license is required.

pip install -r requirements.txt          # click, pyyaml, matplotlib (for tooling)
julia --project=. bootstrap.jl           # resolve + instantiate deps; checks HiGHS
julia --project=. scripts/run_example.jl # ~3 min demo on the synthetic mini-ASEAN

The example solves the core planning model, then the external-pressure story (CBAM €100/t + 24/7 CFE 90% under country-scope vs region-scope import eligibility) and prints a headline cost / CFE / carbon table. Run the full verification suite with julia --project test/runtests.jl (~5 min, all on HiGHS).

scripts/run_example.jl runs on data/mini_asean — a small, fully synthetic 6-country / 12-zone mainland dataset (regenerate any time with python3 tools/make_mini_asean.py). It is the fast, dependency-free way to see the whole pipeline end to end on a laptop before touching the full dataset.

julia --project=. scripts/run_coordination_value.jl   # mini-ASEAN, minutes on a laptop

Solves every country self-sufficient (islanded) and the coordinated regional case on the same dataset, then reports the coordination dividend — Σ(national plans) − coordinated — with per-country detail (cost, installed capacity and carbon intensity, islanded vs coordinated). Point it at the full dataset with a path argument and --method benders for whole-region runs.

Companion engines drill into who benefits, what drives it, and reliability:

julia --project=. scripts/run_coordination_ladder.jl    # per-country benefit at each rung of the ladder
julia --project=. scripts/run_line_attribution.jl       # each corridor's system value + per-country impact (counterfactual)
julia --project=. scripts/run_welfare.jl                # trade-settled benefit per country at zonal LMPs (docs/welfare_settlement.md)
julia --project=. scripts/run_reliability.jl            # existing-fleet adequacy: EUE/LOLE, islanded vs coordinated (docs/reliability.md)

Turn any run's results into a shareable one-file HTML report:

python3 tools/build_report.py --results <results dir> --out report.html   # headline KPIs, mix, per-country, corridors (docs/reporting.md)

The real ASEAN dataset lives in data_asean/{current,2030,2035}/ (all 10 ASEAN members + Timor-Leste, 51 zones). Drive sweeps from a scenario YAML:

python3 scripts/generate_jobs.py -o jobs                     # default 2030 sweep
cd jobs/<name> && julia --project=../.. ../../scripts/run_model.jl

• Scope. Set scope: mainland in a scenario's global_params for continental ASEAN (~26 zones; Borneo + maritime excluded) or scope: full for all 11 countries / 51 zones.
• Single country. Set country: Vietnam (a name from zones.csv) to plan one country islanded.
• Solver. Auto-selected — the open-source HiGHS out of the box, or Gurobi automatically if you have it installed. Force one with solver: highs|gurobi in global_params.

One run = one config.json (see scripts/run_model.jl for keys), validated by the built-in preflight: julia --project scripts/run_model.jl <config> --preflight-only.

Edit a YAML in scenarios/ (e.g. apg_external_pressure.yml): the sweep dimensions (years, coordination levels, external-pressure blocks, in-country RE fractions) and a global_params block of run-wide settings (solver, method, scope, tolerances). No code edits are needed to define a new study.

For large sweeps, scripts/generate_jobs.py writes one job folder per scenario with a config.json and a symlink to scripts/submit_template.sb. Edit the template for your cluster (resources, --mail-user, module load lines), then sbatch each folder or pass --submit. The data path baked into each config.json is absolute, so jobs are location-independent on a shared filesystem.

MIT — see LICENSE. Bundled data carries its own source licenses and attribution requirements; see DATA_SOURCES.md.