Large Language Models for Single-Step and Multi-Step Flight Trajectory Prediction

TL;DR

This work tackles the challenge of predicting aircraft trajectories in ATM by reframing the problem as language modeling with large language models (LLMs). It introduces FTP-LLM, which converts ADS-B trajectory data into language prompts and uses parameter-efficient fine-tuning to train open-source LLMs for single-step and multi-step forecasting, achieving superior accuracy in many settings (notably with LLaMA-3.1) compared to traditional methods. The approach leverages a sliding-window data preprocessing, specialized prompt design, and 4-bit quantization to enable feasible training on commodity hardware, while also highlighting notable latency drawbacks for real-time applications. The results indicate strong potential for LLMs in aviation forecasting, especially in data-limited scenarios, but call for advancements in inference speed and phase-specific robustness to meet operational requirements.

Abstract

Flight trajectory prediction is a critical time series task in aviation. While deep learning methods have shown significant promise, the application of large language models (LLMs) to this domain remains underexplored. This study pioneers the use of LLMs for flight trajectory prediction by reframing it as a language modeling problem. Specifically, We extract features representing the aircraft's position and status from ADS-B flight data to construct a prompt-based dataset, where trajectory waypoints are converted into language tokens. The dataset is then employed to fine-tune LLMs, enabling them to learn complex spatiotemporal patterns for accurate predictions. Comprehensive experiments demonstrate that LLMs achieve notable performance improvements in both single-step and multi-step predictions compared to traditional methods, with LLaMA-3.1 model achieving the highest overall accuracy. However, the high inference latency of LLMs poses a challenge for real-time applications, underscoring the need for further research in this promising direction.

Figures (7)

• Figure 1: Overview of single-step and multi-step prediction in time series tasks.
• Figure 2: Comparison of deep learning-based (e.g. LSTM) and LLM-based methods in time series tasks.
• Figure 3: Overall architecture of the proposed FTP-LLM, comprising data preprocessing, sliding window sampling, prompt construction, fine-tuning on LLMs, and inference for prediction.
• Figure 4: Visualization of the sliding window strategy in single-step prediction.
• Figure 5: An example of prompt template provided to LLMs.