Back to the map

Model coverage and replay accuracy

Power Dispatch Studio is a three-zone economic dispatch model for the Philippine Wholesale Electricity Spot Market (WESM). This page lists the calculations it performs, its replay results against published prices, the available analyst workflows, and its limits.

License and inputs

The code uses the MIT license. Calculated data uses CC BY 4.0. Model inputs come from public records or are labeled assumptions.

Browser and Python calculation

Both interfaces use HiGHS and build the same linear program. A parity test checks the model text and reference results.

Replay results

Recorded days are compared with published prices. Correlation, average error, bias, and high-price-hour agreement are reported below.

Model coverage

Capability	Coverage	Implementation
Economic dispatch, hourly	Yes	A linear program over three zones and 24 hours, priced from the balance duals
Transfer limits between zones	Yes	Two high-voltage direct-current links, with an hourly limit and a wheeling cost
Storage	Yes	State of charge with a round-trip efficiency, and a free terminal state each day
Reserve capacity	Partly	The hourly replay holds the published reserve requirement out of the energy stack. Published reserve prices and a separate offer replay are shown; reserve prices are not co-optimized with energy.
Hydro energy budget	Yes	A daily MWh limit per zone, taken from the recorded day
Shortfall penalty	Yes	Unserved load uses the published ₱32/kWh offer cap. This is a numerical penalty, not an estimate of the value of lost load.
Loss of load probability	Yes	Random forced outages, sampled, reported per zone
Capacity expansion	No	The long-term view reports DOE demand paths and project lists. It does not choose a least-cost build.
Multi-period chronology	Partly	168 hours as one program in one view. Everything else solves one day at a time
Unit commitment	No	An experimental fuel-block minimum-output test is documented on the method page. It is not per-unit commitment and is not a Studio capability.
Named units	Partly	Named plants supply capacity and outage inputs. Energy dispatch remains aggregated by fuel and grid.
Security-constrained nodal dispatch	No	A reduced loss surface over mapped routes, checked per zone. It is not a network model
Contracts and settlement	Partly	A contract book marks against modeled spot, per contract and net. No capacity fee, no wheeling charge, no tax, no credit terms
Stochastic multi-stage	No	Repeated draws over one stage, never a scenario tree

The model dispatches fuel blocks rather than named units. Unit commitment therefore cannot be applied per plant. The method page documents the separate fuel-block experiment and its limits.

Replay accuracy

Every recorded day replays two ways. The cost model prices a competitive fuel-cost floor. The offer replay clears the market operator's own published bid book. Neither reproduces every price, and the table states the gap that remains. Recorded prices never set the model inputs.

Grid	Target	Correlation	Average error	Bias	High-hour hit
Luzon	LWAP	0.73	P2.93	+P1.91	59%
Visayas	LWAP	0.75	P4.77	-P1.25	55%
Mindanao	LWAP	0.81	P4.20	-P1.73	75%
Luzon	MCP	0.75	P2.94	+P2.24	67%
Visayas	MCP	0.77	P5.53	-P4.35	52%
Mindanao	MCP	0.84	P3.74	-P2.88	61%

LWAP is the load-weighted average price and MCP is the market clearing price, both published by the market operator. High-hour hit is the share of the dearest recorded hours the model also calls dear. The nightly build rewrites this table from the current archive, so it cannot freeze behind the data.

Analysis views by task

Analyst task	Studio view
Short-term schedule, one day	Hourly market replay
Supply curve and price-setting fuel	Lowest-cost-first dispatch
N-1 contingency	Power-shortfall risk
Loss of load probability	Power-shortfall risk
PASA, adequacy outlook	Supply after scheduled outages
Price distribution and duration	Price range across recorded days
Capture price by technology	Average price earned by each technology
Interface and transfer limits	Power between island grids
Recorded connection-point differences	Recorded connection-point price differences
Reserve prices and capacity holdback	Reserve market
Demand path and project lists	Annual demand and supply outlook
Market power and HHI	Supplier concentration
Contract position, hedge cover	Contract position
Backcast, model validation	Replay accuracy
Input database	Assumptions and model inputs
Solution file, report	Saved runs, which writes an HTML report and an hourly CSV

Ways to run the project

Map. Review the archive by location.
Studio. Follow an analyst workflow, search the full view library, and share a scenario link.
Python. Install power-dispatch-studio for notebooks and scripts. The package includes a dated archive snapshot for offline use.
Repository. Run make data to rebuild calculated data from the committed source files.

Use other system data

The engine reads dispatch.json and profiles.json from a supplied directory. Those files can describe a different power system.

pip install power-dispatch-studio
power-dispatch run --data-dir ./mysystem --date 2030-01-01

The data contract lists every key those two files must carry, and carries a 30-line system that runs. A test builds that system, solves it, and fails when the document stops matching the engine.

Model limits

The following calculations are outside this model's scope.

No per-unit dispatch or commitment, start cost, or minimum run time.
No full network model. Transmission geometry comes from OpenStreetMap, and the ratings are class defaults.
No multi-year chronology. A year of days is 365 separate 24-hour programs, with storage reset at each midnight.
Three zones, so nothing below the island grid except the reduced loss surface.

The method page explains these limits and identifies the source records, calculations, and assumptions used elsewhere.

Code uses the MIT license. Calculated data products use CC BY 4.0. Source records stay the property of their publishers.