Enhanced Wastewater Treatment Plant Feature Prediction Using Edge Attention Network with Parrot Optimization

To assist process design and controls, enhance system dependability, lower operating costs, and support overall performance optimization, an accurate prediction of wastewater treatment plant (WWTP) key features can help understand and predict plant behavior. In order to address the non-linearity and dynamic character of environmental data, deep learning technologies, which have been shown to be data-driven soft sensors, should be developed for WWTP applications. In order to predict important WWTP parameters such as influent flow, influent temperature, influent biochemical oxygen demand (BOD), effluent chloride, effluent BOD, and power consumption, this work uses deep learning-based models as soft sensors. The proposed method predicts key features of wastewater treatment plants (WWTPs) by using a Multilayer Edge Attention Network with Parrot Optimizer (MEA-Net-PO) and a Hybrid Osprey Optimization Algorithm with Emperor Penguin Optimizer-Based Feature Selection (OOA-EPO). The approach adopts MinMax scaler normalization for preprocessing data to obtain optimal performance in high-dimensional datasets. The MEA-Net-PO model encompasses complex relationships among WWTP data, while OOA-EPO feature selection promotes the selection of the most suitable input features. This does not rely on data distribution assumptions but is very elastic for a given WWTP. The model improves the forecasting efficiency and accuracy. Using historical data from a municipal WWTP, located along a coastal area of Saudi Arabia, the model clearly outperforms traditional methods through RMSE, MAPE, and R2. This may be considered the most promising avenue for data-driven optimization in managing WWTP operations.