摘要
预测电解质分子比的目的是为了控制电解质成分使得电解槽能正常稳定高效运行,并为后续生产、决策提供支持。通过对铝电解产生数据进行分析,研究不同状态下生产规律,针对不同时间窗口的时间序列分子比进行预测:使用改进的人工蜂群算法优化的LSSVM、DeepAR、高斯过程回归算法;并分别进行不同时间窗口的分子比预测,最终将三个预测值放入线性回归模型通过加权计算得到最终预测值。将三个不同时间窗口预测模型放入线性回归中能够较好地预测日分子比值的变化趋势。其中三个独立模型的均方根误差分别为0.29、0.24、0.25,性能优于SVM等传统模型,且最终线性回归预测结果均方根误差为0.25。实际生产中,不同影响因素是否选择不同大小的时间窗口对分子比预测效果达到最优的问题需要进一步分析与研究。将铝电解生产过程中的分子比参数指标及时、准确进行预测,对提交后续生产决策,保证电解槽高效、稳定运行具有重大的现实指导意义。
The purpose of predicting the bath molecular ratio is to control the bath composition,ensure the normal and highly-efficient pot operation,as well as provide support for the subsequent prediction and decision-making process.Through the analysis of data produced by aluminum electrolysis,by studing the production rules under different conditions and by predicting the time series molecular ratio with different time windows:LSSVM improved by ABC algorithm,DeepAR and GPR are used to predict different time windows molecular ratio separatly.Finally,the above three predicted values are put into the linear regression model to get the final predicted value through weighted computation.Three different time window prediction models are put into linear regression,which can better predict the changing trend of daily molecular ratio.The root mean square errors of the three independent models are 0.29,0.24 and 0.25 separately,which are better than SVM and other traditional models,and the root mean square error of the final linear regression prediction result is 0.25.As for the actual production,it needs further analysis and research as for whether different factors choose different size of time window to achieve the best prediction effect of molecular ratio.It is of great practical significance to predict the molecular ratio in the aluminum electrolysis process timely and accurately for submitting subsequent production decisions and ensuring the efficient and stable pot operation.
作者
赵子凌
李晋宏
Zhao Ziling;Li Jinhong(School of Information Science and Technology,North China University of Technology,Beijing 100144,China)
出处
《轻金属》
北大核心
2022年第1期30-39,共10页
Light Metals
关键词
铝电解
分子比
时间序列
aluminium electrolysis
molecular ratio
time series
作者简介
赵子凌(1997-),女,硕士研究生在读,主要研究方向为数据分析与挖掘;通讯作者:李晋宏(1965-),男,教授,博士,主要研究方向为数据挖掘、机器学习、数据可视化。
