Scholarlink

<p>Floods are among the most severe natural disasters, posing significant threats to human lives and property worldwide. Each year, numerous countries suffer substantial casualties and economic losses due to flooding. Therefore, accurate water level prediction is critical for effective flood prevention and disaster mitigation. However, intense rainfall caused by events such as typhoons and tsunamis often results in rapid fluctuations in water levels, making it challenging for traditional machine learning models to capture such nonlinear dynamics, leading to reduced predictive performance. To address this problem, this study proposed an hybrid machine learning model, the VMD-BiLSTM model. The model decomposes complex time series into relatively stable patterns, improving the prediction accuracy, especially when dealing with irregular data. The simulation experiment demonstrated the model’s prediction performance. During testing, the <i>R</i><sup>2</sup> value of the LSTM model was 0.584 and the RMSE value was 0.021&#xa0;m, while the <i>R</i><sup>2</sup> value of the BiLSTM model increased to 0.773 and the RMSE decreased to 0.016&#xa0;m. In contrast, the optimized VMD-BiLSTM model performed better, with an <i>R</i><sup>2</sup> value of 0.940 and a RMSE value of 0.008&#xa0;m. Compared with the LSTM model, the <i>R</i><sup>2</sup> of the VMD-BiLSTM model increased by 61.05% and the RMSE decreased by 61.88%. The outcomes demonstrate that the VMD-BiLSTM model is not only significantly better than the traditional LSTM model in prediction accuracy, but also has better generalization and prediction capabilities. This approach significantly advances flood prediction capabilities, offering practical value for disaster prevention and water resource management applications.</p> Graphical Abstract <p></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

Research on flood prediction based on hybrid AI model VMD_BiLSTM: a case study in BeiJing flood, China

  • 期刊论文

  • 发表日期：

    2026年3月3日

查看全文
  • Ying Nie,
  • Hui Wang,
  • Huaqiang Zhang,
  • Yuqiang Chen

摘要

Floods are among the most severe natural disasters, posing significant threats to human lives and property worldwide. Each year, numerous countries suffer substantial casualties and economic losses due to flooding. Therefore, accurate water level prediction is critical for effective flood prevention and disaster mitigation. However, intense rainfall caused by events such as typhoons and tsunamis often results in rapid fluctuations in water levels, making it challenging for traditional machine learning models to capture such nonlinear dynamics, leading to reduced predictive performance. To address this problem, this study proposed an hybrid machine learning model, the VMD-BiLSTM model. The model decomposes complex time series into relatively stable patterns, improving the prediction accuracy, especially when dealing with irregular data. The simulation experiment demonstrated the model’s prediction performance. During testing, the R2 value of the LSTM model was 0.584 and the RMSE value was 0.021 m, while the R2 value of the BiLSTM model increased to 0.773 and the RMSE decreased to 0.016 m. In contrast, the optimized VMD-BiLSTM model performed better, with an R2 value of 0.940 and a RMSE value of 0.008 m. Compared with the LSTM model, the R2 of the VMD-BiLSTM model increased by 61.05% and the RMSE decreased by 61.88%. The outcomes demonstrate that the VMD-BiLSTM model is not only significantly better than the traditional LSTM model in prediction accuracy, but also has better generalization and prediction capabilities. This approach significantly advances flood prediction capabilities, offering practical value for disaster prevention and water resource management applications.

Graphical Abstract