Generation of synthetic delay time series for air transport applications

TL;DR

This work tackles the problem of generating synthetic, yet realistic, time series of delays at airports, starting from large collections of operations in Europe and the US, and specifically compares three models, two based on state of the art Deep Learning algorithms, and one simplified Genetic Algorithm approach.

Abstract

The generation of synthetic data is receiving increasing attention from the scientific community, thanks to its ability to solve problems like data scarcity and privacy, and is starting to find applications in air transport. We here tackle the problem of generating synthetic, yet realistic, time series of delays at airports, starting from large collections of operations in Europe and the US. We specifically compare three models, two of them based on state of the art Deep Learning algorithms, and one simplified Genetic Algorithm approach. We show how the latter can generate time series that are almost indistinguishable from real ones, while maintaining a high variability. We further validate the resulting time series in a problem of detecting delay propagations between airports. We finally make the synthetic data available to the scientific community.

Figures (13)

• Figure 1: Hyperparameter evaluation for TimeVAE (top) and TimeGAN (bottom). For each combination, 20 synthetic data sets for London Heathrow were generated. Lines show the median discriminative score, and shaded bands indicate the standard deviation. Each subpanel explores a range of values for the corresponding hyperparameter, keeping the others at their optimal values. For TimeVAE, the architecture is limited to three layers of sizes $n \times (1, 2, 4)$, where $n$ is the number of neurons per layer.
• Figure 2: PCA and tSNE visualisations comparing original (red) and synthetic (blue) data sets generated using TimeVAE and TimeGAN, as indicated in the title of the respective figures. Data correspond to arrival delay time series for London Heathrow (EGLL) airport.
• Figure 3: Distribution of classification scores (left panels, blue bars) and of correlations (right panels, cyan bars) for the synthetic data generated with TimeVAE desai2021timevae (left) and TimeGAN yoon2019time (right) for the case of arrival delay time series in Europe. Each bar represents the range between the maximum and minimum of the distribution; the horizontal white line reports the median. See main text for definitions.
• Figure 4: PCA and tSNE visualisations comparing original (red) and synthetic (blue) data sets generated using the simplified Genetic Algorithm approach. Data correspond to arrival delay time series for a single airport, London Heathrow (EGLL).
• Figure 5: Distribution of classification scores (left panels, blue bars) and of correlations (right panels, cyan bars), for the considered airports in Europe (top row) and US (bottom row), and for departure (left column) and arrival (right column) delays. Each bar represents the range between the maximum and minimum of the distribution; the horizontal white line reports the median. See main text for definitions. Full results are further reported in the Appendix.