基于生长神经气改进模糊神经网络的铝电解过程时序数据预测  

Time series data prediction method for aluminum electrolysis process based on GNG-ANFIS

作　　者：盛晓静 吴永明[1,2] 李少波 刘天松[1] 刘应波 SHENG Xiaojing;WU Yongming;LI Shaobo;LIU Tiansong;LIU Yingbo(Key Laboratory of Advanced Manufacturing Technology,Ministry of Education,Guizhou University,Guiyang 550025,China;School of Information,Guizhou University of Finance and Economics,Guiyang 550004,China;Yunnan Institute of Economic and Social Big Data,University of Financeand Economics,Kunming 650221,China)

机构地区：[1]贵州大学现代制造技术教育部重点实验室,贵州贵阳550025 [2]贵州财经大学信息学院,贵州贵阳550004 [3]云南财经大学云南省经济社会大数据研究院,云南昆明650221

出　　处：《计算机集成制造系统》2023年第10期3239-3248,共10页Computer Integrated Manufacturing Systems

基　　金：国家自然科学基金资助项目(51505094);贵州省科学技术基金计划资助项目(ZK[2023]一般079);贵州省科技支撑计划资助项目((2017)2029);云南财经大学科学研究基金资助项目(2020D01)。

摘　　要：针对传统预测模型因分析铝厂时序数据时历史数据量大而无法快速挖掘实时数据隐含的知识信息,导致预测效率低的问题,提出一种基于生长神经气改进模糊神经网络(GNG-ANFIS)全局高效的时序混合预测模型。该模型首先利用生长神经气动态跟踪采集到的时序数据来识别数据奇异点,进而筛选有效数据;然后利用改进后的黑猩猩算法对传统模糊神经网络进行优化;最后,结合铝电解生产过程中铝液杂质铁含量时序数据验证该模型的性能。实验结果表明,混合模型在减少训练时间的情况下仍能准确预测铁含量时序数据,验证了其可行性。The traditional prediction models need to consider a large amount of historical data when analyzing the time series data,and cannot quickly mine the knowledge information implied by real-time data,which causes problems such as low predicting efficiency.To solve the problems,a globally efficient time-series hybrid prediction model based on Growing Neural Gas and improved Adaptive Network-based Fuzzy Inference System(GNG-ANFIS)was proposed.The growing neural gas was used to dynamically track the collected time series data,the identification of data singularities was realized,and then the effective data was filtered.The improved chimp algorithm was used to optimize the traditional fuzzy neural network.The performance of the model was verified by using the time series data of iron content in aluminum liquid during aluminum electrolysis.The experimental results showed that the hybrid model could still accurately predict the iron content time series data while reducing the training time,which verified its feasibility.

关 键 词：铝电解 黑猩猩算法 模糊神经网络 时间序列预测 生长神经气 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]