Tide Level Prediction in Victoria Harbour Based on a VMD-CNN-GRU Combined Model

Subjects: Geosciences >> Marine Sciences Subjects: Mathematics >> Modeling and Simulation submitted time 2024-06-27

Lin Zixuan

Abstract： This study aims to accurately predict the tidal water level height in specific port areas by incorporating meteorological and hydrological factors. We propose a combined prediction model based on Variational Mode Decomposition (VMD), Convolutional Neural Network (CNN), and Gated Recurrent Unit (GRU). Using the public API interface of the Hong Kong Special Administrative Region Government, we collected continuous meteorological and hydrological data from observation points around Victoria Harbour for 12,768 hours from 2023 to 2024. VMD is employed to decompose the original tidal water level time series into multiple sub-modes to reduce sequence complexity. CNN is used to extract features from multi-dimensional historical data, and GRU is utilized to capture the temporal relationships of the features. Compared to traditional time series prediction models such as LSTM, our model shows an average improvement of 64.54 % and 52.40 % in R² and Spearman correlation coefficient, respectively, on the prediction set. RMSE and RAE are reduced by an average of 40.56 % and 32.83 %, respectively, indicating high prediction accuracy. Additionally, the model demonstrates strong robustness through sensitivity analysis. The model also supports flexible adjustment of prediction steps, making it suitable for both long-term and short-term predictions. This provides a reasonable and reliable solution for intelligent monitoring and early warning of tidal water levels in the Greater Bay Area.

Peer Review Status:Awaiting Review

 Hits 469  Downloads 139  Comment 0