A hybrid deep learning framework for predicting industrial wastewater influent quality based on graph optimisation

Due to the impact of excessive influent water quality and water quantity fluctuations, current industrial effluent treatment plants (IETPs) cannot maintain stable decontamination capabilities, posing serious threats to the ecological environment. Therefore, predicting the influent quality of IETPs to facilitate predictive maintenance is imperative. However, because of the highly non-linear nature and variability of industrial wastewater, single models often have limited prediction accuracy. Therefore, this study innovatively constructs a comprehensive and highly adaptable hybrid deep learning prediction framework called GCN optimised CNN-GRU-attention model (GCN-CNNGA). By integrating a graph convolutional network (GCN), convolutional neural network (CNN), gated recurrent unit (GRU), and an attention mechanism, the proposed framework effectively captures the high-dimensional features of influent data and complex non-linear spatiotemporal patterns among variables. Specifically, a CNN is used to accurately extract key local features from the IETP time series. Building on this, a GCN is utilised to further capture the topological information among key influent parameters, revealing their interconnections and deeply mining the complex dependencies among pollutants. The GRU effectively captures long-term dependencies not considered in the above processes, whereas the attention mechanism optimises the weight distribution, thereby enhancing the precision of predictions. During testing using real data from an IETP in an industrial park in East China, the GCN-CNNGA model outperformed the mainstream prediction methods in terms of prediction accuracy and robustness. This study introduces a high-accuracy prediction model that is expected to significantly improve the predictive maintenance levels of IETPs, provide strong technical support for maintaining optimal operational conditions, and have significant implications regarding protecting the ecological environment and promoting sustainable development.