Machine learning-based adaptive large neighborhood search algorithm for the integrated vessel scheduling and speed optimization problem in the compound channel

Abstract

The carbon emission from vessel navigating in the channel accounts for about 61 % of the total emissions in port areas. Considering an effective means of reducing emissions, namely, speed adjustment, this study deals with an integrated problem of vessel scheduling and speed optimization (VSSOP) in the channel. This study considers the complex structure of a compound channel, i.e., containing both one-way and two-way lanes with different navigation rules. We also focus on the effects of meteorological conditions (winds, waves and currents) on the vessel stall, and tidal restrictions on the time window for large vessels to pass through the channel. Thus, a mixed integer programming (MIP) model for the VSSOP is proposed to control the carbon emissions in the channel. Then, we develop a machine learning-based adaptive large neighborhood search (ALNS) approach, where the ALNS is used to solve the proposed MIP in real cases and the dynamic machine learning approach helps to evaluate and fit the complex effects of multiple meteorological conditions on the vessel sailing speed. The dynamic parallel mechanism is further introduced to improve the fitting accuracy of the machine learning part without increasing the running time of the ALNS. The experimental results reveal that the machine learning-based ALNS approach can be applied in practice. Additionally, valuable managerial insights for port operators are obtained to aid in vessel traffic management.