Prediction and optimization of partial denitrification and anammox performance by machine learning and key bioindicators identification.

Optimizing total nitrogen removal efficiency (TNRE) in the partial denitrification and anammox (PDA) system requires a deep understanding of its key driving factors and efficient operational strategies. This study developed a preliminary model for TNRE integrating key parameters and microbial factors (R2 = 0.958). A standalone model with key parameters developed on a larger dataset also demonstrated strong performance with an R2 of 0.921. Feature importance analysis revealed that the TN loading rate, substrate ratios and influent concentration were the primary factors. Moreover, genera Candidatus Jettenia and Denitratisoma were identified as critical bioindicators reflecting system performance. A particle swarm optimization algorithm was employed to develop optimized operational strategy. Implementation of the strategy resulted in a substantial performance improvement and robustness (from 80.70% to 93.52%). These findings provide valuable insights and a framework for optimizing PDA system performance, highlighting the synergistic potential of machine learning and microorganisms in advancing nitrogen removal.