Predicting Port Congestion using Network Analysis

The Problem:

Congestion at US ports has been a growing problem over the course of the COVID19 pandemic. Times for ships to wait in anchorage before unloading or loading their cargo has been growing steadily in almost all US ports, especially in the largest and busiest ports, such as the Port of Los Angeles. These delays can have serious knock-on effects for supply chain and logistics networks, and can have significant economic impacts such as influencing the rise in the inflation rate. There is great value in understanding the drivers of port congestion, and how to best predict it for better logistics and operations planning.

The Data:

We partnered with a provider of AIS data, which consists of vessel transponder signals for most of the globe. These records include position, speed, course, and heading at a frequency up to once every 5 minutes, for over 200,000 vessels daily. We also had access to reference datasets for the ports of the world, as well as attributes about each vessel.

The Solution:

The complexity of this problem required a multi-step approach. First we used the AIS data for stationary vessels to construct geo-fences for each port and anchorage in the world. We then assembled the AIS signals into contiguous voyages for each vessel, as a vessel travelled between ports, anchorages, and lightering zones around the globe. We also used these activities to calculate turn-around times at each location for each vessel. We finally put it all together in a graph network where each port or anchorage is a node, and each edge is a route embedded with aggregate information from the voyages. This graph then used real-time ship behavior across the network to predict turn-around times at the various nodes.

The Impact:

This solution was deployed as part of a maritime logistics planning tool. A user looking to charter a vessel or track a cargo movement engages the software, which uses this network to identify incoming vessels to their origin point, predict ETA to their desired destination, identify the optimal route, and predict the turn-around time at the destination port. This enables the user to better estimate cost and duration of a given voyage, and track progress in real time.