Data-Driven Historical Evaluation and Prediction of Fuel Consumption in Inland Vessels Employing Autonomous Lane Assist
G. Carachino (TU Delft - Mechanical Engineering)
F. Schulte – Mentor (TU Delft - Mechanical Engineering)
M. Saeednia – Mentor (TU Delft - Civil Engineering & Geosciences)
Maryam Pourbeirami Hir – Graduation committee member (TU Delft - Civil Engineering & Geosciences)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
Improving the energy efficiency of inland vessels is an essential step toward meeting European emission-reduction targets. Shipping Technology's Autonomous Lane Assist (ST-ALA) system is designed to automate rudder control and reduce unnecessary steering activity, with the expectation of lowering fuel consumption. However, quantifying its true effect in operational conditions is challenging due to strong confounding arising from vessel characteristics, loading condition, hydrological states, and captain-specific behaviour. This study develops and validates a causal inference framework to estimate the fuel savings attributable to ST-ALA using a large archive of per-traversal data collected by the ST-BRAIN system along the Rhine, >1 million traversals from 132 vessels over 5 years.
A series of models were implemented to estimate the Average Treatment Effect (ATE) of ALA activation on fuel consumption per kilometre. A simple means analysis was used as an unadjusted benchmark. The primary estimator is a machine-learning-based G-computation procedure using a CatBoost decision-tree model to predict counterfactual fuel consumption under toggled ALA status. In parallel, Inverse Probability Weighting (IPW) was applied as a robustness check using a propensity-score model to ensure positivity across relevant covariate strata. Heterogeneous effects were further investigated through Conditional Average Treatment Effect (CATE) aggregation across ships and river environments. Statistical significance was evaluated using influence-function standard errors, bootstrap confidence intervals, and traversal-, edge-, and device-level aggregation schemes.
The estimated global ATE is small: \(0.68\%\) relative fuel savings with a tight confidence interval and statistically significant but economically modest magnitude. CATE analysis confirms consistently small effects across ships and environments, with no subgroup exhibiting large deviations from the global estimate. Furthermore, a pre-departure predictive model was developed using only information available at the start of a voyage. A CatBoost model and a Deep Backpropagation Neural Network (DBPNN) achieve high predictive accuracy and outperform a linear-regression baseline. Model performance was further contextualised through comparison with the GLEC Framework for emissions accounting.
Overall, the results indicate that while ST-ALA has a measurable but modest effect on fuel consumption, the developed methodology provides a robust and generalisable pipeline for causal evaluation and voyage-level fuel prediction in inland shipping.