基于BP神经网络的电化学还原硝酸盐过程智能控制  

Intelligent Control Based on BP Artificial Neural Network for Electrochemical Nitrate Removal

作　　者：张芯婉 孟广源 方立强 常定明 李童 胡锦文 陈鹏[1,4] 刘勇弟[1,4] 张乐华[1,4] Xin-Wan Zhang;Guang-Yuan Meng;Li-Qiang Fang;Ding-Ming Chang;Tong Li;Jin-Wen Hu;Peng Chen;Yong-Di Liu;Le-Hua;Zhang(National Engineering Laboratory of High Concentration Refractory Organic Wastewater Treatment Technology,East China University of Science and Technology,Shanghai 200237,China;Shanghai Espac Environmental Technology Co.,Ltd.,Shanghai 200082,China;Continuing Education Center,Bozhou University,Bozhou 236800,China;State Environmental Protection Key Laboratory of Environmental Risk Assessment and Control on Chemical Process,East China University of Science and Technology,Shanghai 200237,China)

机构地区：[1]华东理工大学高浓度难降解有机废水处理技术国家工程实验室,上海200237 [2]上海易湃富得环保科技有限公司,上海200082 [3]亳州学院教育中心,安徽亳州236800 [4]华东理工大学国家环境保护化工过程环境风险评价与控制重点实验室,上海200237

出　　处：《电化学（中英文）》2023年第12期31-40,共10页Journal of Electrochemistry

基　　金：国家重点研发计划(No.2019YFC0408202);国家自然科学基金(No.21876050)资助。

摘　　要：电化学还原硝酸盐过程关键在于该废水处理过程中参数的有效控制。基于硝态氮电化学还原的测试数据和各参数间的相关性,得出与出水效果密切相关的四因素,即反应时间、初始浓度、初始pH和电流密度,采用BP神经网络算法建立了电化学法还原硝态氮的预测模型,并验证了模型的准确性。结果表明,4-7-1型BP神经网络网络构型最优,模型预测的去除效果与实测值相吻合,R2为0.9095。利用BP神经网络模型对参数调控,可以优化电化学处理过程:对电流密度进行阶段性调控,在相同处理量下可降低15%的能耗;在水质波动情况下进行电流密度控制,在相同处理时间内可保证出水达标。该研究结果可以为智能控制电化学去除硝态氮的过程提供参考。Achieving effective control of parameters in the process of nitrate wastewater treatment is critical to electrochemical water treatment.The powerful nonlinear mapping ability,self-adaptation and self-learning ability of neural network technology can optimize the electrochemical processing.However,there are few researches in this direction.Hence,based on the test data of the electrochemical reduction of nitrate,an electrochemical prediction model was established by using the BP neural network algorithm.Considering the correlation of various parameters in the electrochemical process,the reaction time,initial nitrate nitrogen concentration,pH and current density were determined as the input layer of the BP neural network for model establishment.Results showed that the optimal network configuration of 4-7-1 was achieved by optimizing the hyperparameters of hidden layers number,and the numbers of neurons and epochs.The predicted value of nitrate nitrogen concentration was consistent with the measured value,and the R2 value of 0.9095 was obtained.Meanwhile,the model predicted the effects of initial concentration,pH and current density on the removal efficiency of nitrate nitrogen.In the weak alkaline environment,the stability and reliability of nitrate electroreduction were higher than those in acidic and alkaline environments,and the predicted value of nitrate nitrogen was highly correlated to the true value(R2=0.9908).The initial concentration was negatively correlated to the removal rate,while the current density was positively correlated.Finally,the neural network model was used to control the electrochemical nitrate reduction process.Energy consumption tests were designed by optimizing current density,and15%reduction energy consumption was obtained within the same processing time and processing efficiency.Also,through the prediction model,the effluent quality can be guaranteed by timely adjusting the parameter in the case of sudden water quality changes.The research results can provide a reference for the intellig

关 键 词：硝态氮 电化学还原 BP神经网络 预测模型 智能控制 

分 类 号：X703[环境科学与工程—环境工程] TP183[自动化与计算机技术—控制理论与控制工程]