Author(s): Cristian Camilo Gomez Cortes; Kimberly Solon; Ingmar Nopens; Elena Torfs

Linked Author(s):

Keywords: Global control system; Reinforcement Learning; Liquid Neural Networks; WWTP

Abstract: Wastewater treatment plants (WWTPs) face operational challenges due to their resource-intensive nature, greenhouse gas emissions, and residual sludge production. PID-based control systems struggle with nonlinear disturbances like extreme weather. This study explores Liquid Neural Networks (LNN) -based agent Reinforcement Learning (RL) on a Benchmark Simulation Model No. 2 plant under three scenarios: normal/dry-weather, storm, and winter conditions. The LNN-based agent achieves an average operational cost improvement compared to conventional PID approach, implementing a 70% less connected network than a full-connected approach. The control methodology presented is a promising solution to the complex nonlinear challenges associated with control in WWTPs.

DOI: https://doi.org/10.3850/iahr-hic2483430201-274

Year: 2024