power-dispatch-studio

The PH WESM dispatch engine behind Power Dispatch Studio, as an importable, citable Python module and CLI.

The studio runs the same linear optimization and lowest-cost-first dispatch (merit order) in the browser with HiGHS compiled to WebAssembly. This package is the Python reference code for replaying one recorded day. It builds a supply stack for each island grid from a calculated snapshot of public IEMOP data, applies scenario changes, and clears the three connected grids with inter-grid limits, storage, reserves, and a daily hydro-energy limit. The same calculation and numbers can run in a notebook.

Install

pip install power-dispatch-studio

The only runtime dependency is highspy (the HiGHS solver). A dated snapshot of the public data archive ships in the wheel, so it runs with no network access.

CLI

power-dispatch days                              # recorded days available
power-dispatch run --date 2026-06-15             # hourly CSV to stdout
power-dispatch run --date 2026-06-15 --offer-mode # replay the market's own bids
power-dispatch run --date 2026-06-15 --demand luzon=1500 -o out.csv
power-dispatch run --scenario scenario.json -o out.csv

Python

import power_dispatch as pd

pd.list_days()[:3]
# ['2026-05-01', '2026-05-02', '2026-05-03']

result = pd.run_scenario({
    "date": "2026-06-15",
    "opts": {"demand_delta": {"luzon": 1500}},  # +1.5 GW data-center load
})
result["summary"]["mean_price"]["luzon"]   # PhP/kWh

The scenario override map

A scenario is {"date": "YYYY-MM-DD", "opts": {...}}. The engine accepts the following override keys.

grid is one of luzon, visayas, mindanao.

The scenario file

A scenario carries a schema stamp, so a file written in one place runs in another. The studio writes it, this package reads it, and validate says what is wrong before the solver ever sees it.

power-dispatch validate myscenario.json
power-dispatch validate --keys myscenario.json   # every option, with its meaning
power-dispatch run --scenario myscenario.json -o out.csv

import power_dispatch as pd

pd.validate_scenario(scenario)      # a list of messages, and never raises
pd.load_scenario("myscenario.json") # raises ValueError carrying all of them

Your contract position

A scenario file can carry a contract book, and the engine never reads it. settle marks the book against the modeled spot price, and compare_position reports what a scenario does to it.

pos = pd.compare_position(base, scenario, book, load_mw={"luzon": 400})
pos["net_change_php"]     # the contract gain less the open position's extra cost

It marks energy against modeled spot and stops there. No capacity fee, no wheeling charge, no tax, and no credit terms.

The full key table is in the scenario schema.

The data snapshot

The bundled data is a dated snapshot of the public archive at build time (see power_dispatch/data/meta.json). To run against newer calculated data, point the engine at a copy of the deployed web/data/.

power-dispatch run --date 2026-07-01 --data-dir /path/to/web/data
# or: export POWER_DISPATCH_DATA=/path/to/web/data

That same flag runs your own system. The directory needs dispatch.json and profiles.json, and nothing in it has to be Philippine. The data contract lists every key, with a 30-line system that runs. A test in the repository builds that system, solves it, and fails when the document stops matching the engine.

What this is and is not

This replays recorded days with scenario what-ifs on a documented calculation. It is not a price forecast, and offer mode replays the market's published bids rather than simulating bidding strategy. Recorded market inputs trace to the Independent Electricity Market Operator of the Philippines (IEMOP). Fleet and cost inputs cite their own sources. Caller inputs and model assumptions have separate labels. The Historical replay view and method page report the calculation error. Read the full method.

The engine dispatches fuel blocks per island grid, so it holds no named units and no unit commitment. That is a measured choice. Committing each thermal block with a generic minimum-stable floor lowered the price correlation in all five scored series, from 0.442 to -0.003 in Visayas. pipeline/uc_probe.py in the repository carries the test.

MIT licensed.