Forecast schedule recovery

[mileserickson]

Mukilteo-Clinton ferries typically stay on time in the AM (uni-directional peak volumes eastbound) but get progressively later throughout the afternoon (bi-directional peak volumes). They then gradually return to on-time operation in the late evening. Currently late ferries are predicted to stay late all night.

[anstosa]

Yeah the delay number could be more sophisticated. Here's my proposal for a first step for building a predictive model here:

• Create the model on a nightly basis by pulling all sailing for a given weekly slot for the last 1-2 months (e.g. Saturday at 10:30am or Sunday at 9:30pm)
• Calculate the average delta in departure delays between the previous slot and the current one (e.g. +11mins or -4mins)
• Forecast delays by adding the current actual delay for the vessel to average delay delta for that weekly sailing slot

[mileserickson]

I'd add: a first principles approach might be based on something like: (1) It takes crossing_time to load pedestrians, untie, cross, and dock, and unload pedestrians. (2) Before we can leave on our next trip, we need to unload n_unload cars at a rate of unload_rate per minute. (3) We also need to load min(capacity, n_waiting) cars at a rate of load_rate per minute.

So the forecast additional delay and/or schedule recovery rate can be directly tied to the forecast vehicle loads.