LiuYing,WangXinjie,LiWentao,et al.Machine Learning-based In?situ Monitoring of Specific Energy Consumption in Electrodialysis Processes[J].China Water & Wastewater,2026,42(13):26-34.
Click Copy

Machine Learning-based In?situ Monitoring of Specific Energy Consumption in Electrodialysis Processes

China Water & Wastewater[ISSN:1000-4062/CN:12-1073/TU] volume: 第42卷 Number: 第13期 Page: 26-34 Column: Date of publication: 2026-07-01
Author(s):
Liu　Ying1;  Wang　Xinjie1;  Li　Wentao1;  Zhang　Xinbo1;  Zhao　Yunfeng2;  Chen　Feng3;  Gu　Jinxia4;  Li　Zhimeng5
（1. School of Environmental and Municipal Engineering， Tianjin Chengjian University， Tianjin 300384， China； 2. National Pipe Network Group Northern Pipeline Co. Ltd.， Langfang 065000，China； 3. School of Civil Engineering and Architecture， North China Institute of Aerospace Engineering， Langfang 065000， China； 4. School of Science， Tianjin Chengjian University， Tianjin 300384，China； 5. School of Control and Mechanical Engineering， Tianjin Chengjian University， Tianjin 300384， China）
Keywords:
electrodialysis;  specific energy consumption;  machine learning;  in-situ monitoring;  multilayer perceptron
Abstract:
To address the challenge of online monitoring of specific energy consumption (SEC) in bipolar membrane electrodialysis (BMED) processes, a machine learning (ML)-assisted monitoring approach based on the Internet of Things (IoT) was proposed. Using data obtained from offline prepared NaOH solutions, four predictive models for alkali concentration including multilayer perceptron (MLP), support vector regression (SVR)， gradient boosting regression (GBR) and long short-term memory (LSTM) were compared. The MLP model demonstrated the best performance, achieving a coefficient of determination (R2) of 1.000 0 on the test dataset.The developed model was subsequently transferred for online prediction of alkali concentration during the BMED process. Real-time data were acquired by in-situ sensors deployed within the alkali compartment, which monitored parameters including pH, temperature, conductivity, liquid level, and membrane stack voltage. These data streams enabled the online prediction of alkali concentration and the subsequent derivation of SEC. Experimental results showed that the R2 for online concentration prediction reached 0.905 0, and the derived SEC exhibited a characteristic dynamic evolution pattern of initially rapidly rising to a peak, followed by a decline and minor adjustments, and then slowly increasing over long-term operation. The proposed system integrated a three-tier IoT architecture comprising the perception layer, transmission layer and platform layer, enabling full-process intelligent monitoring that spans from offline model development to online validation. This framework provides a viable and effective tool for facilitating energy-efficient operation and dynamic regulation of the BMED process.
Last Update: 2026-07-01