An improved sparrow search algorithm and CNN-BiLSTM neural network for predicting sea level height.

Area of Science:

• Environmental Science
• Data Science
• Artificial Intelligence

Background:

• Accurate sea level height prediction is crucial for coastal risk management.
• Sea level data presents challenges due to nonlinearity, time-varying, and uncertain characteristics.
• Existing prediction models struggle with the complexity of sea level change data.

Purpose of the Study:

• To develop a more accurate and robust sea level prediction model.
• To optimize the performance of a combined neural network using a novel swarm intelligence algorithm.
• To improve the assessment of sea level risk in coastal areas.

Main Methods:

• A new hybrid algorithm, Sine-Cosine Cauchy Sparrow Search Algorithm (SCSSA), was developed by integrating SSA with sine-cosine and Cauchy variation strategies.
• The SCSSA algorithm was employed to optimize the parameters of a Convolutional Neural Network combined with Bidirectional Long Short-Term Memory (CNN-BiLSTM) neural network.
• The proposed SCSSA-CNN-BiLSTM model was trained and validated using time series data from seven tidal stations in coastal China.

Main Results:

• The SCSSA-CNN-BiLSTM model demonstrated superior performance compared to the standard CNN-BiLSTM model across all evaluation metrics for the SHANWEI Station.
• Across six stations, the SCSSA-CNN-BiLSTM model achieved RMSE values ranging from 20.9217 to 27.8427 mm, MAE from 9.4770 to 17.8603 mm, MAPE from 0.1322% to 0.2482%, and R² from 0.9119 to 0.9759.
• The model consistently provided effective predictions across all tested stations, outperforming other evaluated neural network models.

Conclusions:

• The SCSSA-CNN-BiLSTM model represents a significant advancement in sea level change prediction.
• The proposed model offers high accuracy and robustness, making it a valuable tool for coastal risk assessment.
• Despite increased model complexity, the enhanced prediction capabilities justify its application in critical environmental monitoring.