Optimizing wastewater treatment through combined deep learning and deep reinforcement learning: Recent advances and...

Wastewater treatment plants (WWTPs) are critical components of urban infrastructure, and enhancing their performance while reducing carbon emissions is essential for advancing sustainable urban development. However, WWTPs often face challenges such as fluctuations in water quality and quantity, limitations in real-time monitoring, and delays in operational adjustments, leading to a failure of meeting the discharge standards. The application of artificial intelligence (AI), particularly deep learning (DL) and deep reinforcement learning (DRL), offers significant potential to resolve these complex issues through innovative process optimization. DL has a strong feature extraction capability and data-driven learning paradigm, and can achieve more accurate fitting than DRL when handling tasks involving nonlinear and highly fluctuating process variables; it is particularly effective in fault detection, water quality prediction, and real-time monitoring in wastewater treatment systems. DRL possess a trial-and-error-based learning paradigm, and exhibits greater potential in decision-oriented applications, such as adaptive control of wastewater treatment processes and multi-objective optimization of wastewater treatment plant operations. By clarifying the underlying computational principles of DL and DRL, this review discusses their application suitability and advantages in the wastewater treatment. It emphasizes the need for standardized and open data platforms to support intelligent coupled systems in dynamic and complex wastewater treatment scenarios, and calls for open-source model repositories and the integration of model transparency approaches. Especially, the practical application of intelligent systems in WWTP operation remains challenging, highlighting the need for reliable and standardized data acquisition for real-world applications.