Enhancing Predictive Accuracy of Wastewater Treatment Process: An Approach via Optimizing Data Collection and Increasing Operating State Diversity

Abstract

Excessive nitrogen and phosphorus in wastewater can cause eutrophication, threatening ecology and human health. Wastewater Treatment Plants (WWTPs) are crucial for reducing emissions, but they often operate based on maximum inflow and pollutant concentrations to ensure effluent compliance, leading to energy waste. Machine learning models can predict effluent quality and dynamically adjust WWTP parameters to maintain compliance and reduce energy consumption. However, WWTP data often lacks abnormal states, particularly instances of poor effluent quality, causing data imbalance and low modeling accuracy. In this study, by optimizing system parameters such as the nitrogen loading rate (NLR), hydraulic retention time (HRT), and influent ammonia concentration, a more representative dataset was collected, which significantly improved the prediction accuracy of the models. Optimized data significantly improved model prediction accuracy, with R² values increasing from 0.55-0.63 to above 0.99. Mutual information analysis shows the optimized dataset is rich in information about the Anammox system. This study offers potential methods for stable and energy-efficient wastewater treatment.