Dissolved oxygen (DO) is a critical indicator of water quality in freshwater lake ecosystems. To address the issues of difficulty in prediction of DO, a hybrid model (WT-DSE-LSTM) combined with the wavelet transform algorithm, the dual-squeeze-and-excitation module, and the long short-term memory network is proposed in this paper. The DSE module captures the long-term dependencies and enhances feature weights through the attention mechanism. The MAE, RMSE, and R2 of DO prediction with the proposed model is 0.011, 0.015, and 0.9746, respectively. Furthermore, compared with the state-of-the-art models, the MAE, RMSE of the proposed one can be decreased by 94.09 % and 95.64 % and the R2 of that can be increased by 50.49 %. The DSE module has demonstrated its potential to enhance multivariate time series prediction, which is of great significance for environmental protection and disaster reduction.