Emergency Dispatch Simulator
Simulating dispatch with an LSTM and Voronoi zones
- Year
- 2023
- Kind
- Models
- Status
- Finished
- Built at
- County of Simcoe
A dispatch simulator for the County of Simcoe. An LSTM forecasts where emergency calls will come from, Voronoi zones assign each area to a unit, and the simulator measures response times.
Led a team of three building an emergency dispatch simulator for the County of Simcoe. An LSTM forecasts call volume across the county over space and time. A Voronoi partition, recomputed as conditions change, assigns each area to its nearest unit and guides ambulance and emergency vehicle routing. The simulator measures response times. Requirements came from municipal emergency personnel so it would match real operations.
The problem
When someone calls for an ambulance, how long help takes depends on where units are waiting and which one is sent. A simulator lets you test those choices without touching the real service. This project built one for the County of Simcoe: it forecasts where calls will come from, divides coverage among units and measures the response times that result.
How it works
- 01
Forecast the calls
The simulator starts with demand. An LSTM, a recurrent neural network that carries patterns forward through a sequence, forecasts call volume across the county over space and time: how many calls to expect, and where.
- 02
Split the map into zones
A Voronoi diagram divides a map among a set of points so that every location belongs to the nearest one. With emergency units as the points, each zone is the area a unit covers. The partition here is dynamic, so coverage can be redrawn as conditions change instead of being fixed lines on a map.
- 03
Send the nearest unit
When a simulated call comes in, the zone it lands in decides which unit responds, and that unit is routed to the call. This is how the Voronoi partition guides ambulance and emergency vehicle routing.
- 04
Time every response
The simulator records how long each unit takes to reach its call. Those simulated response times are the measure the optimizer works to bring down.
Decisions
An LSTM for demand
Call volume is a sequence: what happens this hour depends on the hours and days before it. An LSTM is built to remember that kind of history, which makes it a natural fit for forecasting it.
Voronoi for coverage
The rule is easy to state, nearest unit wins, and the partition is quick to recompute. That makes it practical to redraw zones as the situation changes.
Requirements from the people who dispatch
I led a team of three, and we worked through requirements with municipal emergency personnel so the simulator would match how dispatch really works. A simulator that does not match operations produces numbers no one can act on.