Sustainable Maritime Transportation Network considering Sulphur Fuel Regulation - Application of Advanced Machine Learning and Optimization

国际海事组织（海事组织）强调应立即采取行动，减缓碳排放增长，这是气候变化的一个主要因素。联合国贸易和发展会议指出，国际航运承担了世界贸易的80%，对二氧化碳排放量的增加做出了重大贡献。Zis等人（2019年）强调，2015年，许多船舶使用掩体燃料油，占全球硫氧化物排放量的3.5%，从而显著增加了沿海人口稠密地区的环境和健康问题（心脏和肺部疾病）。大气中二氧化碳和硫氧化物浓度的上升导致海洋吸收更多的气体，并变得更加酸性，从而对沿海和海洋生态系统产生重大影响。因此，国际海事组织2020年对排放控制区（ECA）采取了严格的规定，船舶必须使用硫含量不超过0.1%的燃油。此外，为减少海运物流的排放，现有文献强调采取各种措施，如碳税，慢航政策和加油策略。这些措施的采用取决于燃料价格，航运公司面临的一个主要问题是港口（和港口之间）燃料价格的波动。现有的海上物流文献侧重于预测燃料价格仍处于初期阶段，在开发先进的机器学习算法预测掩体燃料价格方面存在空白。该项目将涉及与合作伙伴公司合作，使用包含二氧化碳排放和掩体燃料价格数据的数据集开发用于港口燃料价格预测的机器学习模型。过去的文献忽略了IMO 2020关于使用低硫燃料油加油的规定。因此，本项目将通过开发一个多目标优化模型（混合整数线性规划模型）来最大限度地减少加油成本和排放，从而促进航运公司考虑到海事组织的规定作出加油决策（即选择加油港口和确定加油量）。几位作者强调，需要考虑港口的慢速航行政策（或速度优化）和准确的燃料价格信息，以充分执行掩体燃料管理。因此，目前的研究项目旨在考虑从机器学习模型中获得掩体价格信息，并将其整合到多目标优化模型中，以确定燃料策略，同时最大限度地减少碳和硫排放。以下是指示性的研究问题：1。我们能否开发出一个可靠的预测模型，用于估计掩体燃料价格和港口的二氧化碳排放量？2.我们能否提出一个整体的形式化多目标优化模型（混合整数线性规划模型）来调整海上运输网络？这样的模型需要包含多个目标，包括与可持续性、成本和排放相关的目标，同时捕捉需要调整的合理约束和决策变量。3.我们如何将从预测模型中获得的有关掩体价格信息的见解与优化模型相结合，以确定加油策略，同时促进海上运输网络中的硫和碳减排？4.什么是鲁棒的预测模型和多目标模型来优化问题？

The International Maritime Organization (IMO) highlighted immediate actions to mitigate carbon emission growth, a major factor in climate change. The United Nations Conference on Trade and Development has stated that international shipping carries 80% of world trade making a significant contribution to the rise in carbon dioxide emissions. Zis et al. (2019) highlighted that in 2015, many vessels used bunker fuel oil, which contributes 3.5% to global Sulphur oxides emissions, thereby significantly increasing the environmental and health problems (heart and lung diseases) around populated coastal areas. Rising atmospheric concentrations of carbon dioxide and sulphur oxides are causing oceans to absorb more of the gases and become more acidic leading to a significant impact on coastal and marine ecosystems. Consequently, IMO 2020 adopted strict regulations for emission control areas (ECAs), where ships must use fuel oil with a Sulphur content of no more than 0.1%. Furthermore, for mitigating the emission from maritime logistics, existing literature has highlighted the adoption of various measures such as carbon taxes, slow steaming policy and bunkering strategies. Adoption of such measures depends on fuel prices and a major issue for shipping companies is the fluctuation in fuel prices at (and between) ports.Existing literature on maritime logistics that focuses on predicting fuel prices is still at nascent stage, and there is a gap in the area of developing advanced machine learning algorithms that predict bunker fuel prices. This project will involve working with the partner company, to develop a machine learning model for fuel prices prediction at port using a dataset containing data on CO2 emissions and bunker fuel prices. Past literature has overlooked the IMO 2020 regulations related to the use of low-sulphur fuel oil for bunkering purpose. Therefore, the current project would facilitate bunkering decisions (i.e. choosing the refuelling port and determining the refuelling amount) of the shipping companies considering the IMO regulations by developing a multi-objective optimization model (mixed integer linear programming model) for minimizing the bunkering cost and emissions. Several authors have highlighted the need for considering slow steaming policy (or, speed optimization) and accurate fuel price information at the ports for adequately perform the bunker fuel management. Therefore, the current research project aims to consider bunker price information obtained from the machine learning model and integrate it within the multi-objective optimization model for determining the bunkering strategies while minimizing the carbon and sulphur emissions. The following are indicative research questions:1. Can we develop a reliable predictive model for estimating bunker fuel prices and CO2 emissions at ports?2. Can we propose a holistic formal multi-objective optimization model (mixed integer linear programming model) to tune maritime transportation networks? Such a model would need to comprise several objectives including ones related to sustainability, costs, and emissions, while capturing sensible constraints and decision variables to be tuned. 3. How can we integrate the insights pertaining to bunker price information obtained from the predictive model with the optimisation model for determining the bunkering strategies while facilitating sulphur and carbon emission reduction within the maritime transportation network?4. What would be robust predictive models and multi-objective models to optimize the problem?