Search

With frequent extreme rainfall events caused by rapid changes in the global climate, many cities are threatened by urban flooding. Timely issuance of flood warnings can help prepare for disasters and minimize losses caused by floods. In this study, we propose a method based on a convolutional neural network-bidirectional long short-term memory-difference analysis (CNN-BiLSTM-DA) model for water level prediction analysis and flood warning. The method calculates and analyzes the difference sequence between water level monitoring values and water level prediction values, compares historical flood data to determine the alarm threshold for abnormal water level data, and achieves real-time flood warnings to provide technical references for flood prevention and mitigation. Taking Yancheng city, a low-lying city located in the plain area of Jiangsu Province in China, as an example, this study verifies the accuracy of the CNN-BiLSTM model in water level prediction, which can achieve an accuracy rate above 95%. This provides a reliable data basis for further determination of warning thresholds using the DA model. The CNN-BiLSTM-DA model achieves an accuracy rate of 85.71% in flood warnings without any missed reports, demonstrating that this method has scientific, practical, and accurate features in addressing flood warning issues. [ABSTRACT FROM AUTHOR]