Scholarlink

<p>Accurate weather forecasting can reduce many social and economic losses caused by severe weather. In recent years, spatiotemporal prediction based on deep learning has shown great potential in weather forecasting. However, many existing methods have limited performance in capturing long-term temporal correlations, which makes it difficult to utilize the information at earlier time steps effectively and leads to inaccurate predictions. To address this issue, we develop a Temporal Attention LSTM (TA-LSTM) as the prediction unit by introducing a Temporal Attention Module (TAM) to better utilize the historical information. TAM extends the temporal receptive field of the prediction unit by assigning varying levels of attention to different input information at each time step and accepts all historical information as the input based on the attention levels. In addition, a local attention mechanism is introduced to TA-LSTM to learn spatial correlations from historical data. Experiments on three widely used datasets are performed, and the results demonstrate the superiority of the proposed method, as compared with several state-of-the-art methods.</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

TA-LSTM: Temporal Attention LSTM for spatiotemporal weather prediction

  • 期刊论文

  • 发表日期：

    2025年11月8日

查看全文
  • Jing Dong,
  • Jinxiong Fan,
  • Junzhuo Zhang,
  • Chang Liu,
  • Wei Cheng

摘要

Accurate weather forecasting can reduce many social and economic losses caused by severe weather. In recent years, spatiotemporal prediction based on deep learning has shown great potential in weather forecasting. However, many existing methods have limited performance in capturing long-term temporal correlations, which makes it difficult to utilize the information at earlier time steps effectively and leads to inaccurate predictions. To address this issue, we develop a Temporal Attention LSTM (TA-LSTM) as the prediction unit by introducing a Temporal Attention Module (TAM) to better utilize the historical information. TAM extends the temporal receptive field of the prediction unit by assigning varying levels of attention to different input information at each time step and accepts all historical information as the input based on the attention levels. In addition, a local attention mechanism is introduced to TA-LSTM to learn spatial correlations from historical data. Experiments on three widely used datasets are performed, and the results demonstrate the superiority of the proposed method, as compared with several state-of-the-art methods.