Accurate estimation of tidal level using bidirectional long short-term memory recurrent neural network

• Published in: Ocean engineering Vol. 235; p. 108765
• Main Authors: Bai, Long-Hu, Xu, Hang
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.09.2021
• Subjects: Bi-LSTM model; Multi-step; Short-term estimation; Tidal level; Bi-LSTM model; Tidal level; Short-term estimation; Multi-step
• ISSN: 0029-8018
• DOI: 10.1016/j.oceaneng.2021.108765

The Bi-LSTM model is suggested to evaluate tidal level for short-term. Both single-step and multi-step estimations are taken into account. For the single-step estimation, the measured hourly data from four different tide stations are selected, in which the early 22 hours’ tidal data are chosen as the input and the next hour tidal levels are used as the output. The estimated results by different models are obtained and compared. It is found that the Bi-LSTM model gives the most accurate and stable results. For the multi-step estimation, the measured hourly data in Portland tide station are used, in which the early 22 hours’ tidal data are chosen as the input and the next 24, 48, 96 and 192 h tidal levels are designated as the output respectively. The estimated results by the Bi-LSTM approach are exhibited, which are found to be valid and accurate up to 192 h. Our work shows that the Bi-LSTM model has excellent capability in performing short-term tidal estimations. Few researchers have considered such tidal estimation before. It is expected that this approach can be further used to evaluate other critical time series problems involving in ocean currents and storm surges. •The Bi-LSTM deep neural network model is introduced to forecast tidal levels.•The multi-step forecast model for long-term tide forecast is developed.•The Bi-LSTM model gives the best tidal predictions as compared with traditional ones.•The dependency of traditional models on large quantities of tidal records is overcame.