Maritime Inventory Routing for LNG Distribution from an FSRU Under Carbon Tax: A Column Generation Approach with a Diving Heuristic

A floating storage and regasification unit (FSRU) provides flexible infrastructure for liquefied natural gas (LNG) import, storage, regasification, and regional distribution. Its operation requires coordinated decisions on vessel deployment, sailing speed, loading and discharge timing, terminal inventory, berth use, and backup pipeline supply, while carbon taxation changes the relative cost of alternative vessel and operating choices. This paper formulates this operational planning problem as a maritime inventory routing problem (MIRP) for an LNG distribution network centered on an FSRU. The MIRP formulation integrates FSRU and terminal inventories, heterogeneous vessel capacities, berth constraints, discrete speed choices, pipeline backup, and direct operational carbon costs. To solve the resulting large model, we use a column generation framework with a labeling algorithm for the pricing subproblem, dual stabilization, and a diving heuristic for integer recovery. Each column represents a complete vessel schedule, and the restricted master problem coordinates schedule selection, task coverage, vessel use, backup supply, and aggregate berth capacity. Computational experiments are conducted on calibrated synthetic instances. A complete enumeration benchmark for a small instance provides a diagnostic check, while larger experiments examine computation time, feasible incumbent values, time discretization, instance size, and sensitivity to carbon taxes. The results show nonlinear responses in cost and emissions: low tax rates produce limited operational changes, intermediate rates induce clearer shifts toward vessels with lower emissions at moderate cost increases, and very high stress test rates yield diminishing marginal emission reductions. These findings show how integrated scheduling can support coordinated FSRU planning under carbon costs, while the observed transition points remain specific to the tested instances.