Accurate forecasting of wastewater influent is essential for compliance and load management in wastewater treatment plants (WWTPs). However, conventional single models often show limited accuracy and poor cross-plant transferability due to nonstationarity, heterogeneous sources, and multiscale coupling in influent data. To address these challenges, this study proposes a hybrid forecasting framework integrating Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), the Caterpillar Fungus Optimizer (CFO), and Random Forest (RF), with emphasis on multiplant applicability. Using data from WWTPA, RF was first identified as the optimal base learner, followed by systematic evaluation of seven decomposition methods, through which the CEEMDAN-CFO-RF model was established as most effective. OnWWTPA, the baseline CFO-RF achieved R2 = 0. 78 and MAE = 4937. 55, whereas adding CEEMDAN improved R2 to 0. 96 and reduced MAE and MAPE to 2088. 13 and 1. 62%. Evaluation on two additional WWTPs (B and C), each modeled independently, further suggested that the framework is applicable to WWTPs operating under heterogeneous conditions, yielding R2 ≈ 0. 93 and MAPEs of 2. 69% and 3. 09%, respectively. These results demonstrate the effectiveness and robust generalizability of the proposed CEEMDAN–CFO–RF framework across multiple WWTPs with diverse data characteristics.
Shi et al. (Thu,) studied this question.