Estimating vessel environmental performance : a machine learning approach for predicting vessel fuel consumption and transparently quantifying the environmental sustainability impact of vessel exhaust gases

2020

The shipping industry faces a number of challenges regarding its share of total

anthropogenic emissions worldwide. A range of measures have been initiated, both

by official and private parties. Nevertheless, there are indications that the variety

of approaches and the lack of recognized industry standards are creating confusion,

resulting in ineffective action against the significant problem of climate change. The

growing sense of urgency in relation to global warming, as well as disappointment in the

International Maritime Organizations as the main regulatory body for global shipping,

have created uncertainty and made it more difficult for decision makers to predict the

future of the industry. At the same time, recent advances in data analysis mean that

decision makers are able to produce better empiric modeling of emissions estimations,

potentially improving their operations, regulations, or policies. In this thesis, we propose

a machine learning approach for estimating the environmental performance of vessels.

The theoretical framework established by Gibson et al. (2019) serves as a foundation

for our research. Accordingly, we establish a transparent and quantitative approach for

estimating environmental sustainability impact of vessel exhaust gases. Our approach

uses predicted operational data derived from the Gradient boosting method, together

with the best available measurements of emissions impacts, to portray the complexity of

environmental sustainability. Our findings show the value that empirical modeling, in

the form of machine learning, can provide to internal and external decision makers who

compute and apply emissions estimates, both in the short and long term.

Keywords – Machine learning, Vessel environmental performance, Shipping,

Sustainability