Simulation-Based Ambulance Fleet Allocation

Yisong Yue <>

This is the project page of the Data-Driven Simulation-based Ambulance Allocation project. In particular, we created a simulation of ambulance allocation outcomes that is amenable to allocation optimization. We proposed a simple greedy approach to ambulance allocation that can be shown to be near optimal. More information can be found in [1]. In particular, our simulator is described in Algorithm 1, and our static allocation approach is described in Algorithm 3.

The static simulation dataset is available at This contains all the data necessary for running the static allocation experiments (Section 5 & Section 7.2 in [1]). We sampled 1500 request logs for those experiments. We used the first 500 for training, the next 500 for validation, and final 500 for testing (which is what we reported on).

Each request sample is formatted in 4 lines:
    [request id] [district id] [time of arrival in minutes]
    [priority queue of bases to respond from]
    [ambulance travel time from each base to the scene of the request]
    [total ambulance round trip travel time from base back to base]
For example, in inputs/sample58_week_1.txt, the first 4 lines are:
    0 67 2.33050851551
    39 20 50 30 36
    19.0 19.5764563107 19.9223300971 25.7330097087 27.0550161812
    81.3305085155 81.9069648262 82.2528386126 88.0635182242 89.3855246967
This corresponds to request with request id 0, from district 67, and arriving at 2.33 minutes after simulation begins. There are 5 bases that can service this request, 39, 20, 50, 30, 36. The correspond base-to-scene and total round trip travel times of ambulances sent from each base correspond to the next two lines.

There are 58 bases (labeled 1-58) and 83 districts (labeled 1-83).

A couple of allocation files are given as well. There is the default_allocation.txt, which just assigns one ambulance to each base, and the greedy allocation optimizing cost C1.

The base-to-base travel times is located in base_travel_times.txt. These travel times are given in minutes, and are capped at 30 minutes (i.e., if it takes longer than 30 minutes to travel between bases, we define those two bases to not be traversable). This information was used in our dynamic simulation experiments.

  • [1] Yisong Yue, Lavanya Marla, and Ramayya Krishnan. An Efficient Simulation-based Approach to Ambulance Fleet Allocation and Dynamic Redeployment, In Proceedings of AAAI, 2012 [pdf]

  • [All Content © 2017 Yisong Yue]