CIESC Journal ›› 2024, Vol. 75 ›› Issue (1): 354-365.DOI: 10.11949/0438-1157.20231066

• Process system engineering • Previous Articles     Next Articles

Health state diagnosis of aluminum electrolytic cells based on Adaboost-PSO-SVM

Gang YIN1(), Zhongyou QIAN1, Wenqi CAO2, Pengcheng QUAN2, Hengquan XU2, Feiya YAN3, Min WANG4, Yu XIANG5, Dongmei XIANG6, Jian LU3, Yuhai ZUO7, Wen HE8, Runting LU3   

  1. 1.State Key Laboratory of Coal Mine Disaster Dynamics and Control, College of Resource and Safety Engineering, Chongqing University, Chongqing 400044, China
    2.Aba Aluminium Factory, Aba 623001, Sichuan, China
    3.Guiyang Aluminium Magnesium Design & Research Institute Co. , Ltd. , Guiyang 550081, Guizhou, China
    4.Chongqing Qineng Electric Aluminum Co. , Ltd. , Chongqing 410420, China
    5.Communication NCO Academy, Army Engineering University of PLA, Chongqing 400353, China
    6.China Automobile Research Institute New Energy Technology Co. , Ltd. , Chongqing 400705, China
    7.Qinghai Haiyuan Green Wheel Manufacturing Co. , Ltd. , Xining 810000, Qinghai, China
    8.Bomei Qimingxing Aluminium Co. , Ltd. , Meishan 620010, Sichuan, China
  • Received:2023-10-16 Revised:2023-12-12 Online:2024-03-11 Published:2024-01-25
  • Contact: Gang YIN

基于Adaboost-PSO-SVM的铝电解槽健康状态诊断方法研究

尹刚1(), 钱中友1, 曹文琦2, 全鹏程2, 许亨权2, 颜非亚3, 王民4, 向禹5, 向冬梅6, 卢剑3, 左玉海7, 何文8, 卢润廷3   

  1. 1.煤矿灾害动力学与控制全国重点实验室,重庆大学资源与安全学院,重庆 400044
    2.阿坝铝厂,四川 阿坝 623001
    3.贵阳铝镁设计研究院有限公司,贵州 贵阳 550081
    4.重庆旗能电铝有限公司,重庆 410420
    5.陆军工程大学通信士官学校,重庆 400353
    6.中汽院新能源科技有限公司,重庆 400705
    7.青海海源绿轮制造有限公司,青海 西宁 810000
    8.眉山市博眉启明星铝业有限公司,四川 眉山 620010
  • 通讯作者: 尹刚
  • 作者简介:尹刚(1964—),男,博士,教授,yk115@cqu.edu.cn
  • 基金资助:
    重庆英才·创新创业示范团队项目(CQYC202203091061);科技转化重大项目(H20201555);国家自然科学基金面上项目(62373069)

Abstract:

In order to solve the problem of frequent failures of aluminum electrolytic cells in the aluminum electrolytic production process, a health state diagnosis model of aluminum electrolytic cells based on support vector machine (SVM) was proposed. The thickness of the wall, current efficiency and electrolytic temperature were taken as the comprehensive evaluation indexes of the health state of aluminum electrolytic cells, and the health state of aluminum electrolytic cells was divided into four grades: excellent, good, medium and poor. Considering that traditional support vector machine (SVM) can only be applied to binary classification problem, Adaboost algorithm is used to transform SVM binary classification problem into multi-classification problem to solve aluminum electrolytic cell health diagnosis problem, which fully considers the weight of submodels and strengthens the applicability of the model. The hyperparameters of the model were optimized by using PSO algorithm. The classification accuracy of the model was 94.70% and the Macro-F1 score was 0.9453 in the aluminum electrolytic cells. Compared with the Adaboost-SVM model without optimization algorithm and the PSO-SVM model without integrated algorithm, Adaboost-PSO-SVM improves classification accuracy by 8.34% and 4.93%, and Macro-F1 scores by 8.84% and 5.20%, respectively. Compared with the current mainstream machine learning algorithms DT and KNN, the classification accuracy is improved by 13.64% and 11.11%, respectively, and Macro-F1 scores are improved by 13.47% and 11.04%, respectively. The model provides a comprehensive assessment of the optimal maintenance period for aluminum electrolytic cells. This not only reduces the frequency of failures in aluminum electrolytic cells but also enhances the economic benefits of aluminum plants.

Key words: electrolysis, algorithm, health state, prediction, experimental validation

摘要:

针对铝电解槽在铝电解生产过程中故障频发的问题,提出了一种基于支持向量机(support vector machine, SVM)的铝电解槽健康状态诊断模型,考虑传统的支持向量机只能适用于二分类问题,采用自适应推进算法(adaptive boosting, Adaboost)将支持向量机的二分类问题转化为多分类问题用于求解铝电解槽健康状态诊断问题,充分考虑了子模型的权重,强化了模型的适用性。并利用粒子群优化算法(particle swarm optimization,PSO)对其超参数寻优,提高模型的预测精度。实验结果表明,提出的铝电解槽健康状态诊断模型的准确率和Macro-F1分数分别达到94.70%和0.9453,相较于其他传统模型均有显著提升。

关键词: 电解, 算法, 健康状态, 预测, 实验验证

CLC Number: