ABSTRACT This paper presents an improved stacked ensemble learning framework (AW‐Stacking) that addresses the challenges of multi‐pollutant collaborative prediction and parameter optimization in constructed wetlands. The primary challenges addressed are the limited generalization abilities of models and the absence of a method to resolve conflicts among multiple objectives. The proposed framework integrates diverse learners, including support vector regression, random forest, extreme gradient boosting, and multi‐layer perceptron. It utilizes a dynamic weight allocation matrix along with a hybrid meta‐learning strategy to achieve simultaneous prediction and collaborative optimization of the removal rates of multiple pollutants. Furthermore, this paper introduces an innovative, improved whale‐grey wolf hybrid optimization algorithm (IWOAGWO). This algorithm incorporates adaptive inertia weights and an asymmetric convergence factor, which effectively balances global exploration with local exploitation. It optimally adjusts the configurations of nine key hyperparameters in the model, addressing the oscillation problem that is often caused by traditional random parameterization. Experimental results demonstrate that the IWOAGWO‐AW‐Stacking model achieves an Index of Agreement (IA) of 0.865 for NH 4 + ‐N, 0.838 for NO 2 − ‐N, and 0.822 for total phosphorus (TP), significantly outperforming benchmark models such as random forest and multi‐layer perceptron. Compared to traditional stacking models, the new method shows average improvements in three core indicators: Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and IA, ranging from 20.48% to 27.20%, 10.13% to 18.96%, and 14.92% to 35.18%, respectively. This proposed system extends the use of machine learning in managing water environments and offers a flexible, intelligent decision‐making tool to optimize parameters in artificial wetland systems. It has significant theoretical importance and potential for engineering applications in multi‐scale modeling and the optimization control of complex ecosystems.
Zhang et al. (Sun,) studied this question.