Big Data Helps Estimate Fuel Efficiency

Japan’s Fujitsu Laboratories has developed technology that analyses ship-related big data to estimate fuel efficiency, speed and other performance in actual sea conditions, with less than five percent error. 

The technology utilizes a massive volume of measurement data gathered while the ship is underway, including sensor data of meteorological and hydrographic conditions such as wind, waves, and ocean currents, ship engine log data, and data about the speed and position of the ship. 

By applying the results to Tokyo University of Marine Science and Technology's weather routing simulator and measured data from ships for evaluation, Fujitsu Laboratories demonstrated it could improve fuel efficiency by about five percent from previous results for ships that navigate the shortest shipping routes.

As existing ship performance estimation technologies rely on experiments with model ships in tanks, or on physics model simulations, they could not take into account the complicated interactions of the wind, waves and ocean currents the ship experienced. This led to large margins of error in predictions, says Fujitsu.

With physics models, because physical phenomena, such as the strength of the wind, for example, have to be expressed uniformly in a simplified model, it was impossible to raise the level of estimate accuracy. With this technology, the high-dimensional data, which incorporates a variety of measurement data, is automatically grouped by similar meteorological and hydrographic conditions, and then machine learning and estimation are carried out on each group individually. 

Overly prioritizing actual measured data for machine learning can create a problem where the estimation accuracy goes down for conditions which have not been experienced and there is no measurement data. This problem was solved by automatically adjusting the group boundaries so that no group has data that matches measurement data too closely. This enabled a uniform reduction in prediction error.

The researchers verified that, for a Pacific Ocean shipping route from Tokyo to Los Angeles, by taking an optimal route based on the ship's performance, as determined by the technology, as opposed to the most direct route, fuel consumption could be cut by about five percent, greatly reducing both fuel costs and CO2 emissions. 

Fujitsu states that feeding back data from voyages by previously developed ships into the ship design process, this technology can enable the design of safe ships with high fuel efficiency. In addition, changes in ship performance before and after maintenance and also before and after applying various fuel-efficient technologies can be quantitatively evaluated.
Future Plans

Fujitsu Laboratories will continue to improve prediction accuracy through joint research with Tokyo University of Marine Science and Technology.