Pattern Mining Local Search Heuristics for Dynamic Pickup Delivery Problems

Pickup and delivery is a ubiquitous part of everyday life. Being able to move goods and people between locations in the shortest amount of time and with the fewest vehicles allows packages to arrive on time, allows the just-in-time delivery paradigm that much of our industries rely on, and much more. This paper introduces a new heuristic that uses historical information to improve solutions. This new heuristic is built upon genetic algorithms as a base, as it is flexible and extendable, while being able to solve large scale problems. Previous solutions can be deconstructed into subroutes, with the most frequently occurring being extracted. Using these previously found subroutes, the new heuristic attempts to insert them into newly created solutions. This is done by removing each part of the pattern from the solution, creating route fragments. These fragments can then be recombined to create a new solution. The two principal ideas are that common routes will tend to be good routes and this process allows higher-order operations to be performed.