Autonomous water quality management in an electrochemical desalination process

Area of Science:

• Environmental Science and Engineering
• Materials Science
• Artificial Intelligence

Background:

• Membrane capacitive deionization (MCDI) is a promising technology for water desalination and purification.
• Effective control strategies are crucial for optimizing MCDI system performance and ensuring water quality.
• Existing control methods may struggle with dynamic variations in feed conditions and system fouling.

Purpose of the Study:

• To develop and compare advanced artificial intelligence (AI) control strategies for MCDI systems.
• To evaluate the performance of different artificial neural network (ANN) controllers and reinforcement learning algorithms.
• To optimize control parameters for enhanced water quality and system reliability.

Main Methods:

• Development and evaluation of three ANN controllers: ANN-proportional-integral-derivative, ANN-Integral, and Multiple Parallel ANN-Integral (MPAI).
• Comparison of the MPAI controller with an offline reinforcement learning controller using the Conservative Q-Learning (CQL) algorithm.
• Optimization of the CQL controller using various reward functions, with the Gaussian reward function selected for its effectiveness.

Main Results:

• The MPAI controller demonstrated superior performance compared to other ANN controllers and the CQL controller.
• Both MPAI and CQL controllers effectively maintained effluent concentration around 17 mM despite inlet variations and fouling.
• The MPAI controller achieved higher precision with an error margin near zero, outperforming the CQL controller.

Conclusions:

• AI-driven controllers, particularly the MPAI, offer significant potential for enhancing MCDI system efficiency and reliability.
• Advanced control strategies can robustly manage water quality under challenging operational conditions.
• This research contributes to the advancement of intelligent water treatment technologies.