[Analysis of Genus-level Algal Breakthrough Patterns and Odor Source Tracking in Drinking Water Treatment Processes Using Machine Learning Approach].

To address heterogeneous algal removal efficacy at the genus level and synergistic odor control challenges in drinking water treatment processes， this study conducted comprehensive 8-month （April to November） full-process monitoring at a typical water treatment plant. By integrating random forest regression with SHAP interpretable machine learning， we systematically analyzed removal patterns of 32 algal genera across coagulation-sedimentation， filtration， and disinfection processes while achieving biological source tracking of algae-derived odorants. The results showed that coagulation-sedimentation effectively removed Oscillatoria and Synedra， whereas Microcystis， Planktothrix， and Pseudanabaena exhibited high breakthrough potential. Filtration achieved substantial removal for most genera， though flexible filaments of Lyngbya caused filter penetration. Disinfection efficiently inactivated Oscillatoria and Anabaena but proved ineffective against Planktothrix. Source tracking identified Anabaena as the primary producer of 2-MIB， Oscillatoria for geosmin （GSM）， and Microcystis for β-cyclocitral. This study comprehensively characterizes process-specific algal removal susceptibility， providing a data-driven foundation for dynamic process optimization and odor risk control in water treatment plants.