Optimization of TCN-BiLSTM for dissolved oxygen prediction based on improved sparrow search algorithm

Area of Science:

• Environmental Science
• Water Quality Monitoring
• Ecological Modeling

Background:

• Dissolved oxygen (DO) is critical for river ecosystem health and management.
• Existing DO prediction models face challenges with data noise and feature extraction.

Purpose of the Study:

• To develop an accurate and robust method for dissolved oxygen prediction in river ecosystems.
• To address limitations in current models concerning data noise and dynamic temporal dependencies.

Main Methods:

• Data denoising using the Savitzky-Golay (SG) filter.
• Feature selection via Maximum Information Coefficient (MIC).
• A hybrid Temporal Convolutional Network-Bi-directional Long Short-Term Memory (TCN-BiLSTM) model.
• Hyperparameter optimization using an Improved Sparrow Search Algorithm (ISSA).

Main Results:

• The proposed SMI-TCN BiLSTM model demonstrated superior performance in DO prediction.
• Effective noise reduction and key feature identification were achieved.
• The integrated TCN-BiLSTM module accurately captured temporal dynamics.

Conclusions:

• The SMI-TCN BiLSTM method offers a significant advancement in water quality prediction.
• This approach enhances the protection and sustainable use of river ecosystems.
• The study highlights the effectiveness of combining data preprocessing, hybrid deep learning, and optimized metaheuristics.