基于数据的铝电解槽氧化铝浓度预测  

Prediction of Alumina Density in Aluminum Electrolysis Based on Data

作　　者：崔桂梅[1] 杨海靳 刘丕亮 于凯[2] 

机构地区：[1]内蒙古科技大学信息工程学院,内蒙古包头014010 [2]内蒙古科技大学数理学院,内蒙古包头014010

出　　处：《计算机仿真》2018年第2期305-309,共5页Computer Simulation

基　　金：国家自然科学基金资助项目(61164018);内蒙古自治区自然科学基金资助项目(2014MS0612);内蒙古自治区高等学校科学研究项目(NJZY17163)

摘　　要：根据铝电解的工艺原理和生产数据特点,分析工艺参数对铝电解槽氧化铝浓度的影响。针对铝电解生产数据存在噪声的问题,及铝电解槽在不同槽况下氧化铝浓度不同的特征,提出具有除噪功能FCM算法(NCFCM算法)的槽况分类多支持向量机氧化铝浓度预测方法;上述方法将训练样本数据分为c类,对每个子类样本建立支持向量机预测模型,用粒子群算法优化模型参数;建立判别函数,判别待预测样本数据所属类别;将待预测样本数据代入相应类的回归模型中进行预测。采用某铝厂电解槽采集数据作为应用案例,建立改进方法预测模型。相比标准模糊C-均值聚类算法的支持向量机模型,改进方法不仅考虑多个铝电解工艺参数对氧化铝浓度的影响,且有高预测精度、低训练难度等优点,为铝电解生产过程的稳定提供参考。仿真证实了改进方法的有效性。The influence of process parameters on alumina density is analyzed with the principle of aluminum e- lectrolysis process and characteristics of production data of aluminum electrolysis. Aiming at the noises in the alumi- num electrolysis data and the alumina density difference in different electrolysis cell states, a multi support vector ma- chine model based on Noise - Canceling FCM algorithm （ NCFCM algorithm） is proposed to predict the alumina den- sity. The training sample data were divided into c classes. A SVM was used to build a sub - prediction model with the sample of each subclass. At the same time, a particle swarm optimization algorithm was utilized to optimize the model parameters. Then, discrimination functions were established to recognize which class the sample data belong to. The sample data were put into the regression model of corresponding class to predict alumina density. Used alumi- num electrolysis cell collection data as an application case, the prediction model of improved method was established. Compared with support vector machine model based on standard FCM algorithm, the improved method not only can consider several factors on alumina density, but also has the advantages of high alumina density prediction accuracy and low training difficulty. It can provide reference for the stability of aluminum electrolysis process. The simulation results verify the effectiveness of the improved method.

关 键 词：铝电解 氧化铝浓度 支持向量机 子群优化 回归预测 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]