Consultation: prediction of each vessel's arrival-departure

data1maths

Hello everyone,
I realy need your consultation-help in a certain matter.
I intend to accept my participation in a challenging project. This project is made for the purpose of predicting for each vessel the optimal time of arrival and departure, for one aim which is to optimize the overall congestion of vessels inside the port knowing that it has constant capacity and fluctuating demand that depends on many parameters available from many database servers.
I am wondering if ti is doable to create such a Machine learning\Deep learning model (some suggestions\feedback from you would be very helpful and I’ll be grateful)
I did created many ML models(classification-regression-ensembling models-time series analysis(arima-lstm)……) for many projects before, it(s just that I have never face such challenge and I need some guidance from you sepais  
Thank you in advance.
Best regards.

 

rfuentealba

Hello, @data1maths

 

That is indeed a very challenging project, I envy you!

 

I have tried something similar three years ago, and it was also for vessels in a shipyard. You "can" solve it with Machine Learning if you have optimal entries at first but I would refrain from doing so. Don't go unsupervised, that's my best advice here, unless you want to generate a heatmap on how busy is the shipyard, but once you get your segments it is a long way to interpret these correctly so that you can take action. Your problem is pretty much solved by another area of computer science called heuristics, as it is similar (with certain variations) to the solutions given to the problem of classroom assignment. (Check my wording, as I know the name of that problem in Spanish).

 

You will constantly have a matrix (vessel type, sea level, etc...), and a list of vessels that are heading towards or leaving the shipyard. You can organize a multi-directional cube to select the best combinations of features, and give a vote to them, pondering risks (you don't want to assign a small park to the Titanic, nor try to park the Buque Escuela Esmeralda in a sandy beach). Vessel routes can be calculated by Dijkstra's shortest path algorithm, so that if two routes collide (GIS is your best friend here, so make sure you have a PostgreSQL with PostGIS nearby at any given time) you will have to give priority to the vessel with the largest waiting time to release the pool.

 

I think your architecture will be very similar to the one I talked about in RapidMiner Wisdom, so here is a tip for you:

 

• GeoServer and a local implementation of OpenStreetMap will help you drawing the entire parking/shipyard.
• PostgreSQL+PostGIS will help you with calculations and avoiding collisions, among other things.
• RapidMiner Server will be useful to get the data from operational databases periodically.
• Anaconda Python Distribution has many great mathematical tools that will make your life easier in implementing heuristics such as the one I recommended.
• Leaflet.js or any map viewing application can be connected to your local OSM+GeoServer and an API to help you visualizing how your shipyard is behaving, automatically.

You have a long way to go, my friend.

Hope this helps,

 

Rodrigo.