Related Experiment Video
Updated: Jun 20, 2025

Ion-Exchange Membranes for the Fabrication of Reverse Electrodialysis Device
Published on: July 20, 2021
Autonomous real-time control for membrane capacitive deionization.
Jaegyu Shim1, Suin Lee2, Nakyeong Yun3
1Department of Civil Urban Earth and Environmental Engineering, Ulsan National Institute of Science and Technology, UNIST-gil 50, Ulsan 44919, Republic of Korea.
This study introduces a reinforcement learning (RL) control for membrane capacitive deionization (MCDI) to optimize energy efficiency. The actor-critic (A2C) agent achieved superior performance, significantly reducing energy consumption during ion separation.
Area of Science:
- Water treatment technologies
- Environmental engineering
- Artificial intelligence applications
Background:
- Membrane capacitive deionization (MCDI) is an emerging ion separation technology.
- Optimizing MCDI performance for energy efficiency is crucial but challenging.
- Real-time control strategies for MCDI have not been extensively investigated.
Purpose of the Study:
- To develop a reinforcement learning (RL)-based control model for MCDI.
- To investigate energy-efficient operation strategies for MCDI using RL.
- To minimize both outflow concentration and energy consumption in MCDI.
Main Methods:
- Developed three long-short term memory (LSTM) models for predicting applied voltage, outflow pH, and electrical conductivity.
- Trained four RL agents to optimize MCDI operation.
- Utilized actor-critic (A2C) and proximal policy optimization (PPO2) agents for control.
- Employed Shapley additive explanation (SHAP) to interpret A2C decision-making.
Main Results:
- A2C and PPO2 agents successfully achieved the ion separation goal (<0.8 mS/cm) by reducing electrical current and pump speed.
- A2C demonstrated significantly lower energy consumption (0.0128 kWh/m³ ) compared to PPO2 (0.0363 kWh/m³).
- SHAP analysis provided insights into the influence of input parameters on A2C's control decisions.
Conclusions:
- Reinforcement learning-based control is feasible for optimizing MCDI operations.
- The developed RL model can enhance the energy efficiency of water treatment.
- This approach holds potential for future advancements in water purification technologies.
Related Concept Videos
MOS Capacitor
The metal gate is typically made from highly conductive materials such as aluminum or polysilicon. Beneath the metal gate lies a thin layer of...
Potentiometry: Membrane Electrodes
Controlled-Potential Coulometry: Electrolytic Methods
The chosen potential...
Capacitors
When a voltage source is connected to a capacitor, positive and negative charges accumulate on the opposite plates. This accumulation generates a potential difference that equals the product of the...
Dialysis
Controlled-Current Coulometry: Overview

