Enhancing Civil Aviation Day-to-Day Passenger and Cargo Volume Forecasting with iTransformer: Multimodal Integration and Predictive Accuracy

This paper introduces and enhances the iTransformer model, a prediction framework based on the Transformer architecture, designed to forecast passenger flow and cargo volume on Chinese civil aviation routes. As the civil aviation industry rapidly evolves, accurate predictions of future passenger and cargo volumes are critical for optimizing resource allocation and improving operational efficiency. Drawing on the success of the Transformer model in natural language processing and Large Language Models, iTransformer utilizes self-attention and multi-head attention mechanisms to capture long-term dependencies in time series data. To cater to the unique characteristics of aviation data, iTransformer incorporates several improvements. It features a multimodal fusion mechanism that integrates historical passenger and cargo volumes with external factors like weather, holidays, and special events, enhancing predictive capability. Additionally, a time embedding layer in the encoder and decoder allows the model to better understand periodic and trend variations, improving prediction accuracy. The loss function includes weighted mean squared error (MSE) and mean absolute error (MAE), balancing performance across routes with different traffic levels. Using a real civil aviation dataset with multiple international and domestic routes over several years, iTransformer outperforms traditional time series models and recent deep learning models in accuracy and stability. Specifically, the model maintains a prediction error within 4.5% for a three-day forecast, 6.63% for a ten-day forecast, and approximately 7.79% for a fifteen-day forecast, as demonstrated in our experimental results. Practical applications of iTransformer, such as forecasting passenger flow trends for better flight scheduling and predicting cargo volume trends for optimized resource allocation in logistics, validate its feasibility. Furthermore, potential applications in public transportation flow prediction and electricity demand forecasting are explored. Future research could further improve the model’s structure, data preprocessing methods, and loss function design, enhancing its performance across various scenarios. This study showcases the effectiveness and superiority of iTransformer in predicting passenger flow and cargo volume in civil aviation, providing new technical tools and theoretical support for route planning and capacity allocation.