Scholarlink

<p>As air traffic volume increases, air traffic management (ATM) becomes increasingly important. Accordingly, accurate aircraft trajectory prediction (ATP) is essential for estimating and mitigating potential conflict risks. Among various ATP approaches, deep learning methods are adopted in this study. For sequential trajectory data, a sequence-to-sequence long short-term memory (LSTM) model and a Transformer model are considered. These models are trained using a teacher forcing approach to leverage the ground truth (GT) output sequence during training and to alleviate the significant computational loads of autoregressive training. To address the issue of output error accumulation during inference, an incremental-standard-deviation-based (ISTD) noise model is proposed and incorporated into the decoder input of the models. Numerical simulations based on both synthetic and real airspace scenarios are conducted to evaluate the effectiveness of the proposed approach.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Incremental-Noise-Based Teacher Forcing for Sequence Learning in Aircraft Trajectory Prediction

  • 期刊论文

  • 发表日期：

    2026年1月28日

查看全文
  • Hyunjin Choi,
  • Somang Lee,
  • Yeonju Eun

摘要

As air traffic volume increases, air traffic management (ATM) becomes increasingly important. Accordingly, accurate aircraft trajectory prediction (ATP) is essential for estimating and mitigating potential conflict risks. Among various ATP approaches, deep learning methods are adopted in this study. For sequential trajectory data, a sequence-to-sequence long short-term memory (LSTM) model and a Transformer model are considered. These models are trained using a teacher forcing approach to leverage the ground truth (GT) output sequence during training and to alleviate the significant computational loads of autoregressive training. To address the issue of output error accumulation during inference, an incremental-standard-deviation-based (ISTD) noise model is proposed and incorporated into the decoder input of the models. Numerical simulations based on both synthetic and real airspace scenarios are conducted to evaluate the effectiveness of the proposed approach.