Data-driven jet fuel demand forecasting: A case study of Copenhagen Airport

TL;DR

This study addresses short-term jet fuel demand forecasting for the Danish market by comparing data-driven approaches—SARIMA, Prophet, LSTM, and a SARIMA–LSTM hybrid—across three real-world case studies using daily data from a major fuel distributor to forecast 30 days ahead. It finds that traditional time-series methods (SARIMA) often perform best, though LSTM can outperform in volatile contexts, and the usefulness of exogenous variables hinges on data quality; in the CPH airport case, SARIMAX with exogenous inputs achieves the strongest accuracy. Overall, the work highlights the value and limits of data-driven forecasting for aviation fuel sourcing, demonstrating improved decision support for suppliers and airlines through case-specific model selection and careful data handling.

Abstract

Accurate forecasting of jet fuel demand is crucial for optimizing supply chain operations in the aviation market. Fuel distributors specifically require precise estimates to avoid inventory shortages or excesses. However, there is a lack of studies that analyze the jet fuel demand forecasting problem using machine learning models. Instead, many industry practitioners rely on deterministic or expertise-based models. In this research, we evaluate the performance of data-driven approaches using a substantial amount of data obtained from a major aviation fuel distributor in the Danish market. Our analysis compares the predictive capabilities of traditional time series models, Prophet, LSTM sequence-to-sequence neural networks, and hybrid models. A key challenge in developing these models is the required forecasting horizon, as fuel demand needs to be predicted for the next 30 days to optimize sourcing strategies. To ensure the reliability of the data-driven approaches and provide valuable insights to practitioners, we analyze three different datasets. The primary objective of this study is to present a comprehensive case study on jet fuel demand forecasting, demonstrating the advantages of employing data-driven models and highlighting the impact of incorporating additional variables in the predictive models.

Figures (10)

• Figure 1: An LSTM cell and its components: $h_t$ represents the hidden state, $c_t$ is the cell state and $x_t$ is the input value at time $t$.
• Figure 2: Schema of the SARIMAX-LSTM hybrid model.
• Figure 3: Scaled time series for the first case study, reporting the actual amount of fuel bought, the flight schedule of the same carrier that was available one month before, and the overall number of flights expected to depart from CPH airport.
• Figure 4: Test set predictions obtained with the SARIMA model on the first case study. The dashed red line represents the mean prediction and the shaded region corresponds to the 5% confidence interval.
• Figure 5: Predicting the test set residuals of the SARIMA model with a univariate LSTM model on the first case study. The dashed orange line represents the mean prediction and the shaded region corresponds to the 5% confidence interval.