Abstract: Dissolved oxygen (DO) is a pivotal indicator in river water quality monitoring. In order to accurately predict the DO concentration in river water bodies, we developed a VMD-GA-BiLSTM deep learning integration model, combining Variational Mode Decomposition (VMD), Genetic Algorithm (GA), and Bidirectional Long Short-term Memory (BiLSTM). We also conducted training and testing on DO concentration data from Aixinzhuang section in Xingtai City for the years 2020-2023, with comparisons made against the prediction results of multiple classic deep learning models (BiLSTM, GA-BiLSTM and EMD-GA-BiLSTM). The results revealed that the VMD-GA-BiLSTM model achieved remarkable performance on the test set, with RMSE of 0.149, MAE of 0.135, and R2 of 0.974. When compared to BiLSTM, GA-BiLSTM and EMD-GA-BiLSTM models, the model demonstrated significant improvements: RMSE was reduced by 0.464, 0.307, and 0.290, respectively; MAE was decreased by 0.413, 0.173, and 0.239, respectively; and R2 was increased by 0.216, 0.133, and 0.088, respectively. These findings demonstrated the superior prediction accuracy of the constructed model. To further validate the versatility and stability of the model, we applied it to predict three water quality indicators: pH, DO, and ammonia nitrogen, in Houxiwuqiao section of Xingtai City. The results indicated that the VMD-GA-BiLSTM model outperformed other classical models, achieving the lowest RMSE and MAE values and the highest R2 score. This demonstrated the model's high adaptability and robustness in predicting water quality time series data. The research results showed that the VMD-GA-BiLSTM model can accurately predict concentrations of DO and other water quality indicators. This model can serve as a scientific basis for the sustainable utilization of water resources.