Evaluating the influence of marine weather parameters uncertainties on the ship fuel consumption with Monte Carlo analysis

Abstract

This study analyzes the impact of weather parameter uncertainties on ship fuel consumption using Monte Carlo simulations. A feed-forward neural network (FFNN) is trained on ship and weather data to predict fuel use. The voyage route is discretized, and ensemble weather data from ECMWF ERA5 (1940-2024) are collected for each point. Probability distributions are fitted to these variables, and randomized scenarios are generated. The generated FFNN model is then used to simulate fuel consumption under varying conditions, and the resulting uncertainties are assessed using statistical metrics such as standard deviation, confidence intervals, and density plots. The generated FFNN models achieved high predictive accuracy, with MAE ranging between 0.6065 and 0.7240 kg & sdot;min-1 and MAPE from 0.9743% to 1.1690%, with R2 = 0.99. The goodness-of-fit analysis of the weather variables reveals that the Lognormal distribution provides the best fit for most variables based on log-likelihood, AIC, and BIC criteria. In addition, the analysis highlights that fuel consumption variability is closely tied to changing weather conditions along the route, with higher standard deviations indicating unstable fuel usage due to environmental fluctuations, while lower values reflect more consistent and stable operating conditions.