Modeling Method for A²O Anoxic Zone Based on PSO-SCN

This study proposes a particle swarm optimization-optimized stochastic configuration network (PSO-SCN) model for the anoxic zone of the A²O wastewater treatment process by integrating mechanistic analysis with data-driven modeling. Data processing involves the use of isolation forest for outlier handling, KNN for missing values, and KPCA for dimensionality reduction (from seven to four dimensions). Validated with data from an A²O simulator and an actual wastewater treatment plant, the model demonstrates superior accuracy (RMSE and NSE) in predicting effluent COD, NH₄ ⁺-N, and NO₃ ^--N concentrations compared to the unoptimized SCN model, the traditional mechanistic model (ASM1), and other classical models such as PSO-BP and PSO-RBF. Furthermore, SHAP analysis enhances the model's interpretability. The results indicate that the PSO-SCN framework achieves an effective balance among prediction accuracy, computational efficiency, and mechanistic interpretability, providing a valuable basic tool for intelligent wastewater treatment control.