文献类型：期刊文献

中文题名：基于神经网络精准预测含铬废酸电氧化再生效果

英文题名：Accurate prediction for electro-oxidation regeneration of chromiumcontaining waste acid based on artificial neural network

作者：师雅琪[1];孟广源[1];陈鹏[2];张芯婉[1];付涛[1];杨正武[1];张连胜[3];张乐华[1,4,5]

机构：[1]华东理工大学工业废水无害化与资源化国家工程研究中心,上海200237;[2]合肥大学生物食品与环境学院,安徽合肥230601;[3]黑龙江广盛达新材料科技有限公司,黑龙江哈尔滨158100;[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;[5]石河子大学化学化工学院环境监测与污染物控制重点实验室,新疆石河子832003

年份：2025

卷号：45

期号：1

起止页码：131

中文期刊名：工业水处理

外文期刊名：Industrial Water Treatment

收录：;北大核心:【北大核心2023】；

基金：2023年鸡西市石墨产业揭榜挂帅项目(JKJB2023H03)。

语种：中文

中文关键词：隔膜体系;电氧化;含铬废酸;人工神经网络;资源化再生

外文关键词：membrane system;electro-oxidation;chromium containing waste acid;artificial neural network;re?source regeneration

摘要：“铬法”膨胀石墨生产过程中产生的废酸具有酸浓度大、铬含量高等特征,可以采用隔膜体系通过电氧化法将Cr(Ⅲ)氧化为Cr(Ⅵ),实现含铬废酸的资源化再生。基于该强酸体系难以实现Cr(Ⅵ)含量的实时检测,开展了基于神经网络精准预测含铬废酸电氧化再生效果的研究。在含铬废酸再生实验基础上,首先采用相关性分析方法确定了电解时间、H_(2)SO_(4)浓度和电解液体积为Cr(Ⅵ)再生的关键特征参数,然后通过超参数优化获得人工神经网络的相对最优拓扑结构:神经元数量=35、批训练样本数=30、隐藏层层数=4,构建模型预测值与实验值的决定系数(R^(2))大于0.97,均方根误差(RMSE)小于0.04。最后经实验验证,模型预测值与实验值的平均相对误差最大为0.14%,表明模型具有很好的泛化能力。人工神经网络模型克服了由于多参数、非线性与时变性造成的电化学过程预测难的问题,可以实现复杂映射条件下对Cr(Ⅵ)再生的预测,对电化学过程的优化调控具有重要意义。
The waste acid generated in the production of expanded graphite by“chromium method”h as the charac?teristics of high acid concentration and high chromium content.The Cr(Ⅲ)can be oxidized to Cr(Ⅵ)by electrooxidation in membrane system to realize the regeneration of waste acid containing chromium.Since it was difficult to achieve real-time detection of Cr(Ⅵ)content in this highly acidic system,a study based on artificial neural network was conducted to accurately predict the electro-oxidation regeneration effect of chromium-containing waste acid.Based on the regeneration of chromium-containing waste acid experiments,the key characteristic parameters of hexavalent chromium regeneration including time,sulfuric acid concentration,and electrolyte volume were deter?mined by correlation analysis.Then,through hyperparameter optimization,the relatively optimal topology structure of the artificial neural network was obtained as follows:Neurons=35,Batch size=30,Layers=4.The coefficient of de?termination(R^(2))between predicted value and experimental value was greater than 0.97,and the root-mean-square er?ror(RMSE)was less than 0.04.Finally,the average relative error between predicted value and experimental value was 0.14%,which indicated that the model had good generalization ability.The artificial neural network model over?came the difficulty of predicting electrochemical processes due to multi-parameter,nonlinearity and time variability,and could realize the prediction of Cr(Ⅵ)regeneration under complex mapping conditions,which was of great signifi?cance for the optimization and control of electrochemical processes.