A Lot-Based VMD-CNN-BIGRU Indoor Mariculture Water Quality Prediction Method Including Attention Mechanism

The water quality in indoor mariculture systems directly impacts the health of aquaculture species. Dissolved oxygen (DO) is a crucial indicator in these systems, and accurate, rapid prediction of DO levels is essential for improving the yield and quality of aquaculture products. This study addresses the challenge of complex DO variations and low prediction accuracy in indoor aquaculture environments. We propose a hybrid prediction model combining Variational Mode Decomposition (VMD), Convolutional Neural Network (CNN), and Bidirectional Gated Recurrent Unit (BiGRU) with an attention mechanism. The model utilizes water quality data from grouper aquaculture facilities. The VMD algorithm enhances data stability by decomposing non-smooth signals into multiple non-frequency-scaled subsequences. These subsequences are then trained and predicted using the combined model. We conducted comparative experiments with four models: BiGRU, CNN, CNN-BiGRU, and VMD-CNN-BiGRU with an attention mechanism. The VMD-CNN-BiGRU-Attention model demonstrated superior performance, with a root mean square error (RMSE) of 0.0306, a mean absolute error (MAE) of 0.0252, and a coefficient of determination ([Formula: see text]) of 0.9817. Our findings indicate that the VMD-CNN-BiGRU-Attention model meets the predictive needs for DO in indoor mariculture, providing a reliable reference for monitoring and early warning control of water quality in aquaculture systems.