Fast-Time Simulation: Visualisation & Making Decisions
I have only just touched the surface of this concept, and I am excited by its power, but it is time to wrap up this series on fast-time simulation. I’ve created a flight schedule, built an airport and subjected it to almost three years of punishment. Today’s post is a quick look at visualisation of the results and what decisions it could support.
As I wrote yesterday, analysis needs a purpose. Like any research, it needs a research question. And the clearer that question is, the easier it is to devise a methodology and determine an answer. So, although this might seem a little backward, given I’ve already built the model and run the simulation, let’s go with this:
Based on the forecast mean busy day, should we increase the number of apron parking positions?
And let’s assume that there are currently five stands.
Visualisation
Fast-time simulation can produce immense amounts of data. For example, each of these simulations saw 88 arrivals and 88 departures. They each had scheduled and actual times for those events, plus whether they went into overflow and what stand they were parked on (plus a time for this, if not parked from landing). From this, we can derive delay times, on-time performance; the list goes on.
Therefore, data visualisation is essential but still a challenge. An effective visualisation requires familiarity with the data and the research question. Like any use of statistics, selecting the wrong variables can tell a different story (whether on purpose or not).
In line with the what-if basis of our simulation and the research question above, I chose to present the super-simulation results on a per-stand number basis. The diagram below shows the results for each subset of simulations based on the randomly determined number of stands. Each graph contains around 125 simulations and shows how the airport in that configuration performs across a range of chaos scenarios.
Click to make bigger
This diagram shows that our airport will struggle with five stands if this forecast traffic schedule is realised. We can expect upwards of ten aircraft waiting for parking even on the good days. If weather or something else impacts flights, then that number rises.
Obviously, ten stands would be great. It looks calm through all chaos scenarios. But it is unlikely that any airport will invest in a project to double its apron capacity to ensure minimal impact in all scenarios. Business decisions are always a balancing act and risk taking care of the uncertainties. So, a much more realistic assessment is working out where on the scale from six to nine, we get value out of our investment.
Six stands offers a noticeable jump down in overflow at low chaos values. It still gets messy when things go pear-shaped, but we could live with this.
Seven stands looks to make another stepped improvement for low chaos and holds on to this performance for longer when chaos increases.
I don’t see much improvement when we get to eight stands at low chaos, but it looks a little nicer around chaos levels nine to eleven.
There is a lot of similarity between nine stands and ten, so this might be overdoing it too.
Without really knowing what chaos means, I am hard-pressed to justify a specific choice. I would probably support a case for two additional stands to be built. I am always confident in the operations team to deal with anything thrown their way. So we could make it work!
ChatGPT Prologue
So, how did ChatGPT help me through this process?
Well, it got me started, and I helped with lots of little code questions throughout. But it wasn’t a case of explaining what I wanted, and it gave me the answer. In fact, it gave me incorrect code a lot of the time. I would tell it about the error I got, and then it would apologise and give me new code. Sometimes, that wouldn’t work, either!
It was like having a Year 12 Computer Science student helping me out. They could help with the easy stuff but fall apart when the question got too niche or complicated.
