Time series prediction of airport operational resilience under severe weather conditions

Abstract

Airport operational resilience is a crucial metric reflecting an airport’s capacity to adapt to external shocks, essential for maintaining safety and operational efficiency. While there has been research on airport resilience under various severe weather conditions, the specific contributing factors and their impacts remain inadequately explored. This study develops a comprehensive index system that integrates airport performance and meteorological data, using a random forest algorithm to quantify the influence of various factors on airport resilience across five types of severe weather. Furthermore, a PatchTST(Patch time series Transformer)-based time series model improved by the Cauchy loss function is proposed to accurately predict airport operational resilience. Focusing on severe weather events during the period from January 2023 to July 2024 at Dallas-Fort Worth International Airport in the United States. To mitigate multicollinearity, variables with high Pearson correlation and variance inflation factor (VIF) values were removed prior to analysis. Feature importance results reveal that hourly flight movements (HFM) consistently hold the highest importance across weather types, while temperature (TEMP), relative humidity (RHUM) and air pressure (PRES) exhibit relatively higher meteorological influence despite limited overall impact. The optimal Cauchy-PatchTST model, with a look-back window of $L=36$ and a forecast length of $T=1$, outperforms the traditional PatchTST model with MSE loss, three Transformer-based models and other optimized machine learning algorithms, achieving a 15.49% to 94.10% reduction in MAE on the test set. This study provides critical indicator analysis for airports across various severe weather conditions and offers reliable resilience data to support future operations management.