Or, to ask that question in a different way, why is operational advice from a machine learning based system so accurate? And what does this mean in practice?
What is machine learning?
In machine learning, we (1) take some vessel operational data, (2) train a performance model of the vessel using that data, and (3), use the trained model to make operational predictions based on new vessel data.
After the initial training, the performance model enters an on-going learning mode where it’s exposed to new, unfamiliar data step by step. At each step, the model makes operational predictions and gets feedback about how accurate they were. This input is used to enhance the performance model so that future predictions (based on more new data) are more accurate.
GreenSteam Discover uses historical vessel data (1-2 years’ worth of operational, voyage and fuel bunkering data) to build a performance baseline model of the vessel. Once this is in place, it can use new, real-time vessel data to calculate inefficiencies and make predictions on how the vessel will perform in the future.
It is the ability of our machine learning system to make sense of the myriad combinations of weather, sea-state, loading, fuel calorific value etc. that enable it to very accurately predict how the vessel will perform in situations in which the trained performance model has incomplete, or no, data.
Alternative technologies have to make assumptions about how a vessel will perform in new situations for which no data is available, and this approach simply results in less accurate predictions, and this can have very negative consequences.
Alternative technologies have to make a large number of assumptions about how a vessel will perform in situations that have not previously been encountered. This is the single biggest advantage of a machine learning based system. The predictive advice that the machine learning system provides isn’t based on any assumptions – it’s all derived from the trained performance model that is itself based on real-world experiences.
What does it mean though when we say that the advice coming from a machine learning-derived performance model of a vessel is more accurate?
It might seem obvious, but the more accurate the advice, the greater the likelihood that the vessel will indeed be performing at optimum efficiency and therefore maximising fuel savings.
For example, the implementation of trim settings advice that has a a +/- 10% error level, could result in the vessel operating in a less favourable configuration than if they crew had not followed it. In other words, inaccurate advice will likely result in the vessel performing even less efficiently – meaning wasted fuel. This is why accurate advice is central to optimising operational efficiency.
Only an artificial intelligence-based machine learning approach can make sense of the vast number of permutations of vessel parameters and external forces acting on the vessel. Making highly accurate predictions about the optimal way to operate a vessel requires a performance model that learns as it goes along, and that thrives (and becomes ever more accurate) the more data that’s thrown at it.