A Multi-Step Prediction Model Based on Wavelet Transform and Deep Learning Techniques for Predicting Dissolved Oxygen Concentration

Dissolved oxygen (DO) is a critical indicator of water quality in freshwater lake ecosystems. Due to the nonlinear and non-stationary characteristics of dissolved oxygen, which are influenced by multiple factors and fluctuate with seasons, achieving high-precision predictions is challenging, whether in the short or medium to long term. To address this issue, we propose a new hybrid model for multi-step prediction of water quality in advance, which combined with the Wavelet Transform algorithm, the Dual-Stage-Squeeze-and-Excitation module, and the Long Short-Term Memory (WT-DSE-LSTM) network. The Wavelet transform technology reduces interference noise in water quality data and the DSE module captures the long-term dependencies and enhances feature weights through a two-level attention mechanism, then using a two-layer LSTM network to output prediction results based on historical information. We use the WT-DSE-LSTM model to predict DO in Beijing-Hangzhou Grand Canal in Hangzhou with a MAE of 0.007 and an RMSE of 0.018. Our experiment proves that this model is a feasible multi-step water quality prediction method, which is of great significance for environmental protection and disaster reduction.