Capesize Voyage Speed Optimization: A Reinforcement Learning Model for Fuel, Earnings, and CII
Developed by Veson Nautical in collaboration with Bayes Business School, this whitepaper examines how reinforcement learning could help optimise Capesize vessel speeds. Using data from more than 130,000 voyages, it considers the trade-offs between fuel consumption, earnings and CII performance.
“Introduction
Voyage planning has grown considerably more complex in recent decades. Operators must simultaneously balance fluctuating bunker fuel prices, dynamic freight markets, port congestion, charter party obligations, and increasingly stringent emission regulations. Voyage optimisation determines both commercial viability and now environmental compliance. For operators of Capesize Bulkers, the stakes are high. The combination of long voyage distances and high bunker consumption means that even marginal speed optimisation can yield substantial economic gains and emissions reductions. This analysis presents the results of collaborative research between Veson Nautical and Bayes Business School, in which we developed and evaluated a deep reinforcement learning model for Capesize Bulker speed optimisation. The research draws on 130,388 voyages that took place between July 2020 and June 2026, taken from VesselsValue’s Trade AIS tracking. The model is trained on the voyages that took place before April 2025 and tested on voyages that followed, comparing its speed recommendations with the actual speeds sailed as observed in the data. The model improved on operator decisions for fuel, earnings, and carbon intensity. On the 18,443 voyages used for testing, the model cut fuel consumption by an average of 47.2 metric tonnes per voyage (9.98%) and improved economic performance by $42,156 per voyage (22.6% improvement). Additionally, the speeds recommended by the model increased the share of annual emissions ratings that would score “A” under the IMO’s Carbon Intensity Indicator (CII) from 76.1% to 90.5%. This paper sets out how the model works, what we found, and what the results could mean for owners, charterers, and regulators.”
Kasinis, M. and Delf, P. (2026) Capesize Voyage Speed Optimization: A Reinforcement Learning Model for Fuel, Earnings, and CII. Veson Nautical and Bayes Business School. Available at: Veson Nautical (Accessed: 16 November 2026).
The full white paper is available to read via the download button above, or through the Veson Nautical website.
To read other Clean Maritime Research Hub outputs, please see: Resources – UK National Clean Maritime Research Hub