Multi-input multi-output temporal convolutional network for predicting the long-term water quality of ocean ranches

Area of Science:

• Marine environmental monitoring
• Aquaculture science
• Predictive modeling

Background:

• Marine water quality prediction is vital for ocean ranching stability and aquatic life.
• Existing models struggle with complex, dynamic ocean environments, leading to inaccuracies and poor long-term predictability.

Purpose of the Study:

• To develop an advanced prediction model for marine water quality parameters.
• To address the limitations of current methods in accuracy and complexity for ocean ranching environments.

Main Methods:

• Proposed a Multi-Input Multi-Output Temporal Convolutional Network (MIMO-TCN) model.
• Utilized ConvNeXt for feature extraction and TCN for prediction, incorporating skip connections to enhance performance.
• Evaluated model robustness and prediction accuracy using dissolved oxygen data from ocean pastures.

Main Results:

• The MIMO-TCN model demonstrated superior prediction accuracy compared to other learning models.
• Achieved average reductions in Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE) of 60.77%, 30.88%, and 52.45%, respectively.
• Showcased an average improvement of 6.07% in R-squared value, indicating enhanced predictive power.

Conclusions:

• The proposed MIMO-TCN model offers a significant advancement in marine water quality forecasting.
• The method provides a reliable basis for scientific decision-making in marine environment control and aquaculture management.