Artificial Intelligence-Empowered Ecological Wetland System: A New Model for Nitrogen and Phosphorus Pollution Control in Lakes—A Case Study of Lake Dianchi

Lake Dianchi, the largest freshwater lake on the Yunnan-Guizhou Plateau, has suffered from severe eutrophication due to excessive nitrogen (N) and phosphorus (P) inputs. Conventional constructed wetlands show large seasonal efficiency fluctuations and lagged management responses. Here we propose a four-layer AI-wetland framework (perception, analysis, decision, execution) and validate it at the 383-ha Hai-Hong wetland. An XGBoost-SHAP model predicted 7-day N/P removal rates with ±10% error; a Delft3D-based digital twin was coupled to NSGA-II for multi-objective optimisation. Compared with traditional static operation, AI-driven dynamic control increased annual total N and total P removal by 15% and 15%, respectively, and reduced cash OPEX by 22% (Table 2). During a 50-year storm event, the system lowered peak outflow TP by 30% through preemptive water-level drawdown. Although upfront CAPEX (≈ CNY 18–22 million) was financed by green bonds, annual major-repair reserves (2% of CAPEX) and ICT electricity must be budgeted. The study provides a replicable, cost-effective pathway toward SDG 6.