A PCA-EEMD-CNN-Attention-GRU-Encoder-Decoder Accurate Prediction Model for Key Parameters of Seawater Quality in...

Area of Science:

• Environmental Science
• Data Science
• Machine Learning

Background:

• Seawater quality prediction accuracy is often limited.
• Effective prediction models are crucial for aquaculture management and environmental monitoring.
• Existing models face challenges in handling complex water quality data.

Purpose of the Study:

• To develop an optimized water quality parameter prediction model with enhanced accuracy.
• To improve the prediction of key water quality parameters in marine environments.
• To provide a reliable decision-making basis for aquaculture management in Zhanjiang Bay.

Main Methods:

• Principal Component Analysis (PCA) for key factor screening.
• Ensemble Empirical Mode Decomposition (EEMD) for data denoising.
• A 2D-CNN module for feature extraction.
• An integrated attention-GRU-encoder-decoder (attention-GED) module for prediction.

Main Results:

• The PCA-EEMD-CNN-attention-GED model achieved high accuracy in both short-term and long-term predictions.
• Short-term prediction yielded RMSE of 0.246, MAPE of 0.307, and R² of 97.80%.
• Long-term prediction yielded RMSE of 0.878, MAPE of 0.594, and R² of 92.23%, outperforming baseline models.

Conclusions:

• The proposed PCA-EEMD-CNN-attention-GED model significantly improves seawater quality prediction accuracy.
• The model offers a robust solution for predicting key water quality parameters.
• It provides valuable insights for water quality control and aquaculture management in Zhanjiang Bay.