Enhanced Prediction of Dissolved Oxygen Concentration using a Hybrid Deep Learning Approach with Sinusoidal Geometric Mode Decomposition

Dissolved Oxygen (DO) is a crucial indicator in water bodies, enabling assessment of eutrophication degree, ecosystem status, self-purification capacity, and water quality health. This paper presents a hybrid model for predicting DO. The model utilizes symplectic geometric mode decomposition (SGMD) to decompose the DO data into multiple intrinsic mode functions (IMF) components, allowing extraction of trend and seasonal information. Subsequently, a hybrid deep learning model based on convolutional neural network (CNN) and Temporal Convolutional Network (TCN) is constructed to predict each IMF component and reconstruct the predicted value of DO. Comparative analysis with other benchmark models demonstrates the superior accuracy, indicating its effectiveness in DO prediction. Furthermore, the study investigates the impact of incorporating different water quality variables on DO prediction accuracy, revealing that incorporating variables with high correlation enhances accuracy. The accurate prediction of DO concentration by the SGMG-CNN-TCN model holds promise for sustainable river water management and plays a significant role in optimizing water environment management.