Self-labelling of tugboat operation using unsupervised machine learning and intensity indicator
The digital data related to the tugboat's actual operation, such as a time sequence of fuel consumption and speed, is usually unlabeled or untagged to a particular activity such as tugging or cruising. There is a great interest in identifying the type of operation for a particular data point for tugboats. The operation type is an operational label for every data point that is essential for further analytics, as operations for tugging and cruising use different fuel and navigation profiles. Therefore, it is important to eventually handle them separately. This paper aims to develop a self-labelling framework for tugboat operation by using unsupervised machine learning and a proposed intensity indicator by considering two sets of data, namely: the positional data and the fuel consumption rate data. The fuel consumption data are actual data obtained from mass flowmeters installed on tugboats whereas the positional data are navigational data purchased from marine data aggregators. The developed self-labelling enables ship operators in identifying operations and locations that require heavy fuel consumption. It can also be used for further big data analytics and machine learning for fuel consumption prediction when given vessel speeds.
Journal/Conference/Book titleMaritime Transport Research, Special Issue: Big Data and AI
- 6524 (R-MOE-A403-C002) Ship's Energy Efficiency via Machine Learning with Big Data Analytics