Optimizing Feeder Network Design with Deep Reinforcement Learning: A Hyperheuristic Approach

Abstract

BORA : Yes Across the globe, hundreds of shipping networks form an intricate web of trade routes forming the backbone of international commerce. These networks are responsible for an estimated 80 percent of all cargo transported globally and are known as Liner Shipping Network Design Problem (LSNDP) in the literature. This thesis will focus on a variant of the LSNDP known as the feeder networks. It is the problem of serving a number of shipping requests using a fleet of vessels. Each request involves moving a number of containers from the origin port to the destination port. Our objective is to design routes that connect all ports in the most optimized order such that pickup and deliveries correspond with the lowest cost possible. We will implement, adapt and compare two state-of-the-art frameworks, where one (Adaptive Heuristic) framework is optimized and created for the FNDP while the other (Deep Reinforcement Learning Hyperheuristic) is a more general framework for a multitude of different Combinatorial Optimization Problems.