Logistics - Supply chain
Ocean freight companies handle millions of containers spread all over the world on hundreds of ships, hundreds of ports, thousands of depots and hundreds of thousands of customers. They need to be able to provide empty containers to customers who need them and to ensure that enough empty containers are available at the right time in the right place. It might be necessary to dispatch empty containers to meet this demand.
The problem is therefore to predict when and where empty containers will be available.
Addressing this requires to handle problems of incompleteness (missing information) and of uncertainty. Indeed, as there is no system for locating containers, their exact locations are unknown: only some observation points are available when they pass some checkpoints.
Besides, the time it takes for a container to travel from A to B is very uncertain and depends on a multitude of variables. For example, the time it spends in customs is highly variable and depends on factors such as national or local holidays or political and social events. The time it spends on land at a customer’s premises to be either loaded or unloaded is also highly variable and uncertain.
The probabilistic solution to this problem is to evaluate the different possible trajectories (routes with transit times) for each container. It is obviously impossible to evaluate all of them, so we have to sample all of these possibilities.
This sampling remains extremely expensive in terms of computation time and having dedicated hardware increasing the performance by several orders of magnitude allows to be more accurate by sampling more solutions, and to make these predictions more often, which ultimately allows serving more customers and reducing operating costs.