Abstract:

In wastewater treatment processes, efficient modeling of key water quality-related parameters is crucial for achieving process control optimization, anomaly detection, and decision support. However, process data typically exhibit complex temporal dependencies, multivariable coupling nexus, and non-stationary characteristics, which pose significant challenges for modeling. To address these issues, this paper proposes a Lightweight Multi-Parameter Time Series Prediction Method for Wastewater Treatment based on the Smooth Wavelet Transform (SWT) and Collaborative Attention (CA). This method first performs multi-scale decomposition on wastewater data and uses the stationary wavelet transform to extract features from the data at different scales. Then, it constructs a collaborative attention mechanism based on geometric attention and sparse attention to efficiently capture the complex coupling relationships and temporal features among key water quality parameters. Finally, a dual projection layer maps the reconstructed features obtained from the inverse wavelet transform to the final prediction outputs. The model is trained and evaluated using real-world data collected from a wastewater treatment plant in Dongguan. Experimental results show that, in 12-step forecasting tasks, the proposed model achieves a reduction of 9.15% to 37.70% in RMSSD compared to baseline models. In other tasks, its accuracy is second only to TimesNet, which has a significantly larger parameter size. These findings demonstrate the model’s strong balance between lightweight design and predictive performance, validating its effectiveness for time series forecasting in wastewater treatment.

Key words: multivariate time series prediction, stationary wavelet transform, collaborative attention mechanism, lightweight, wastewater treatment