← Back to work

Simulation · 2026

Elevator Simulation

A discrete-event simulation answering a real capital question: how many elevators does a five-floor commercial building actually need — one, two, or three?

elevator configurations compared
1–3
waiting-time percentiles, not just means
P95
realistic time-varying demand
NHPP

Architecture

Select a stage to explore it.

01 · Passenger demand

Arrivals follow a non-homogeneous Poisson process — morning rush, lunch spikes, evening exodus — because uniform demand would make every configuration look fine.

SimPy 4course project — BDA450, Seneca

Visuals

Elevator Simulation GitHub repository with discrete-event simulation documentation
GitHub repository — SimPy discrete-event simulation comparing elevator configurations

01 · Problem

Elevator capacity is a one-time capital decision with a permanent service consequence. Undersize and tenants wait every day; oversize and the building paid for steel it didn't need. Intuition can't answer this — queueing dynamics are famously counterintuitive.

02 · Solution

A SimPy discrete-event simulation of a five-floor commercial building under realistic time-varying demand, comparing one, two, and three elevators under a cooperative SCAN policy, and reporting the full waiting-time distribution for each.

03 · Engineering decisions

Demand is non-homogeneous on purpose — peak-hour behavior is what sizing decisions are about, and uniform arrivals would hide it. Percentiles and extreme delays are first-class outputs, because a mean wait of 40 seconds is meaningless to the person who waited six minutes.

04 · Results

A quantified cost-versus-service frontier for the building: how much waiting time each additional elevator buys, including its effect on worst-case delays and cross-floor fairness.

05 · Lessons learned

Simulation is judgment infrastructure: it doesn't make the decision, it makes the trade-offs visible. And distributions beat averages every time someone's experience lives in the tail.