A Water Quality Prediction Method Based on the Deep LSTM Network Considering Correlation in Smart Mariculture

Area of Science:

• Marine biology
• Aquaculture technology
• Data science

Background:

• Cage-cultured water quality prediction is crucial for smart mariculture.
• Environmental factors introduce nonlinearity, dynamism, and complexity in water quality parameters.
• Traditional forecasting methods suffer from low accuracy, poor generalization, and high time complexity.

Purpose of the Study:

• To propose a novel water quality prediction method using deep long short-term memory (LSTM) learning networks.
• To accurately predict key water quality parameters like pH and water temperature in mariculture settings.
• To address the shortcomings of traditional forecasting methods in terms of accuracy and efficiency.

Main Methods:

• Data preprocessing including linear interpolation, smoothing, and moving average filtering.
• Utilizing Pearson's correlation coefficient to identify relationships between water quality parameters.
• Constructing a water quality prediction model based on LSTM with preprocessed data and correlation information.

Main Results:

• Short-term prediction accuracy for pH reached 98.56% and for water temperature reached 98.97%.
• Short-term prediction time costs were 0.273 s for pH and 0.257 s for water temperature.
• Long-term prediction accuracy for pH reached 95.76% and for water temperature reached 96.88%.

Conclusions:

• The proposed LSTM-based model significantly improves the accuracy of short-term and long-term water quality prediction.
• The method effectively handles the complex, dynamic, and nonlinear nature of mariculture environments.
• This approach offers a viable solution for smart mariculture by enhancing water quality forecasting capabilities.