《海洋预报》| 基于VMD-CNN-BiGRU-Attention的澳门地区城市淹没水位预测模型研究

摘要：澳门地区由于地理位置特殊、气候湿润以及高度城市化的特点，在台风风暴潮期间频繁遭受由降水、河道洪水、倒灌海水等多因素叠加引发的复合型洪涝事件。为了精准分析这类灾害下的城市淹没水位的变化规律，提出一种基于变分模态分解（VMD）、卷积神经网络（CNN）、双向门控循环单元（BiGRU）和注意力机制的城市淹没水位预测模型。首先，采用 VMD对历史城市淹没水位数据进行分解，得到一系列相对平稳的子序列；然后，将CNN用于对多环境因素数据进行特征提取；特征提取完成后，使用BiGRU网络进行双向循环训练；最后，通过注意力机制为BiGRU输出分配相应权重，并加权求和得到最终的城市淹没水位预测结果。实验结果表明，该模型在城市淹没水位变化预测中表现优异。

关键词：风暴潮 城市淹没水位预测 变分模态分解 卷积神经网络 双向门控制循环单元 注意力机制 

Abstract：Due to its geographical location, humid climate and high degree of urbanization, the Macao region frequently suffers from compound flooding events caused by the superposition of multiple factors such as precipitation, river floods and backflow of seawater during typhoon storm surges. To accurately analyze the changes in urban inundation water levels under such disasters, a prediction model for urban inundation water levels based on Variational Mode Decomposition (VMD), Convolutional Neural Networks (CNN), Bidirectional Gated Recurrent Unit (BiGRU) and attention mechanism is proposed. Firstly, the VMD is used to decompose historical urban inundation water level data to obtain a series of relatively stable sub-sequences. Then, the CNN is used to extract features from multi-environmental factor data. After feature extraction, the BiGRU network is used for bidirectional recurrent training. Finally, the attention mechanism assigns corresponding weights to the BiGRU and performs weighted summation to obtain the final prediction result of urban inundation water levels. Experimental results show that this model performs well in predicting changes in urban inundation water levels.

Key words：storm surge; urban inundation water level prediction; variational modal decomposition; convolutional neural networks; Two-way door control circulation unit; Adaptive attention mechanism

读者也可以通过公众号菜单栏

“相关知识”板块进入

《当期目录》和《过刊浏览》

查阅详细期刊内容

未来两天海洋环境预报

35个美丽海岛环境预报