Both mechanical models and machine learning-based models are widely utilized for real-time dynamic control
however, their implementation in the water sector often incurs significant data and computational costs. To address these challenges, this study introduces an innovative feature extraction method designed to enhance the cost-effectiveness of dynamic control in wastewater treatment plants. The proposed method extracts dynamic features from time-series data of key substrate variables to construct a data-driven model and develop real-time control strategies. The results indicate that the data-driven model accurately predicts the variation trends of ammonia nitrogen, total nitrogen, and biochemical oxygen demand, with correlation coefficients exceeding 0.8. Compared to the traditional activated sludge 2D model, the proposed approach significantly improves computational efficiency, reducing model parameter calibration time from 939.75 s to 87.52 s. Furthermore, the developed real-time control strategies reduce energy consumption by up to 24.3% while ensuring effluent quality meets discharge standards. The inclusion of a dynamic update mechanism, which refreshes model parameters every three hours, further enhances system adaptability and responsiveness. In conclusion, the proposed method minimizes reliance on complex water quality, sludge, and environmental datasets by directly extracting dynamic biochemical characteristics from key variables, providing a cost-effective solution for dynamic control in wastewater management.