基于MLP神经网络的铝电解槽出铝量预测  

Prediction of aluminum output in aluminum electrolytic cell based on MLP neural network

作　　者：倪小峰 曹斌 NI Xiaofeng;CAO Bin(College of Big Date and Information Engineering,Guizhou University,Guiyang 550025,China;Chinalco Intelligent(Hangzhou)Safety Science Research Institute Co.Ltd,Hangzhou 310000,China)

机构地区：[1]贵州大学大数据与信息工程学院,贵阳550025 [2]中铝智能科技发展有限公司,杭州310000

出　　处：《智能计算机与应用》2021年第8期139-142,共4页Intelligent Computer and Applications

摘　　要：铝电解生产是一个大延迟、多变量耦合和非线性的过程。铝电解槽每天的出铝量往往是根据多年积累的经验制定的,目前为止还没有一个准确的计算方案。根据电解槽的各个状态参数进行相关性分析,其中电解质水平、铝水平、槽温、分子比变化等一些列参数对出铝量的影响相对较大。本文将从操控系统得到的各参数数据集分为测试集、训练集,利用机器学习框架搭建MLP神经网络日出铝量模型,通过模型得到预测值和真实值进行对比,对误差曲线进行了分析。Aluminum electrolysis production is a process with large delay,multivate coupling and nonlinear.The daily aluminum output of aluminum electrolytic cell is often based on years of accumulated experience,so for there has not been an accurate scheme.According to the state parameters of the electrolytic cell,the correlation analysis shows that some parameters,such as electrolyte level,cell temperature and molecular ratio change,have a relatively great influence on the amount of aluminum output,each parameter data set obtained data set obtained from the control system is divided into test set and training set.The daily aluminum quantity model of MLP neural network was built by machine learning framework,then the predicted value and real value were compared through the model,and the error curve was analyzed.

关 键 词：MLP 训练集 测试集 预测值 

分 类 号：TF821[冶金工程—有色金属冶金]