基于PID搜索优化的CNN-LSTM-Attention铝电解槽电解温度预测方法研究  

Research on CNN-LSTM-Attention aluminum electrolyzer electrolysistemperature prediction method based on PID search optimization

作　　者：尹刚[1,2] 朱淼 全鹏程 颜玥涵 刘期烈[5] Yin Gang;Zhu Miao;Quan Pengcheng;Yan Yuehan;Liu Qilie(State Key Laboratory of Coal Mine Disaster Dynamics and Control,Chongqing University,Chongqing 400044,China;College of Resource and Safety Engineering,Chongqing University,Chongqing 400044,China;Aba Aluminium Factory,Aba 623001,China;School of Electrical Engineering,Guizhou University,Guiyang 550025,China;Chongqing Post and Communications University,Chongqing 400065,China)

机构地区：[1]重庆大学煤矿灾害动力学与控制全国重点实验室,重庆400044 [2]重庆大学资源与安全学院,重庆400044 [3]阿坝铝厂,阿坝623001 [4]贵州大学电气工程学院,贵阳550025 [5]重庆邮电大学,重庆400065

出　　处：《仪器仪表学报》2025年第1期324-337,共14页Chinese Journal of Scientific Instrument

基　　金：重庆市自然科学基金创新发展联合基金项目(CSTB2024NSCQ-LZX0166);重庆英才·创新创业示范团队项目(cstc2024ycjh-bgzxm0131);科技转化重大项目(H20201555)资助。

摘　　要：铝电解生产环境恶劣,受电场、磁场、流场、温度场等多物理场耦合影响,导致铝电解生产过程故障频发。铝电解温度是影响铝电解槽寿命和运行状态的重要参数,但由于槽内温度很高且具有强烈腐蚀性,至今尚未找到有效的电解温度在线检测与预测方法。为了解决这一技术难题,通过理论分析结合现场实验验证,揭示了铝电解槽电解温度与其工艺参数间的密切相关性,并据此提出一种基于深度学习的铝电解槽电解温度预测模型。考虑到铝电解槽工艺参数的复杂性、非线性、高维度、时序性等特征,采用卷积神经网络(CNN)用于提取数据的高维特征,长短期记忆网络用于建模(LSTM),处理铝电解生产过程中的时序数据,引入了注意力机制(Attention),学习输入参数不同部分之间的关联性,同时根据输入数据的重要程度进行加权处理,并采用PID搜索优化算法(PSA)对CNN-LSTM-Attention模型的参数进行寻优,减少训练时间并提高模型的性能。最后经铝电解实际生产数据进行现场实验验证,结果表明:提出的温度预测模型相关指数(R~2)为0.963 7,均方根误差(RMSE)和平均绝对误差(MAE)分别为5.417 6和3.382 5,与单一模型算法、其他预测算法和不同优化算法对比验证表明,该模型的性能更佳,能够准确预测铝电解槽电解温度,实现了铝电解槽电解温度的在线检测。Aluminum electrolysis production environment is harsh,affected by electric field,magnetic field,flow field,temperature field and other multi-physical field coupling,resulting in frequent failures in the aluminum electrolysis production process.Aluminum electrolysis temperature is an important parameter that affects the life and operation status of aluminum electrolysis tanks,but due to the high temperature and strong corrosiveness of the tank,no effective online detection and prediction method for electrolysis temperature has been found so far.In order to solve this technical problem,the close correlation between the electrolysis temperature of aluminum electrolyzer and its process parameters is revealed through theoretical analysis combined with on-site experimental validation,and a deep learning-based prediction model for the electrolysis temperature of aluminum electrolyzer is proposed accordingly.Considering the complexity,nonlinearity,high dimensionality,and temporal sequence of the process parameters of aluminum electrolyzer,Convolutional Neural Networks(CNN)is used to extract the high-dimensional features of the data,and Long Short-Term Memory networks(LSTM)is used for modeling.The Attention mechanism is introduced to learn the correlation between different parts of the input parameters while weighting the input data according to its importance,and the PID-based search algorithm(PSA)is used to optimize the CNN-Attention model for the aluminum electrolysis production process.The PID-based search algorithm(PSA)is used to optimize the parameters of the LSTM-Attention model to reduce the training time and improve the performance of the model.The results show that the proposed temperature prediction model has a correlation index(R 2)of 0.9637,root mean square error(RMSE)and mean absolute error(MAE)of 5.4176 and 3.3825,respectively.Comparison of the proposed temperature prediction model with a single-model algorithm,other prediction algorithms,and different optimization algorithms shows that the model performs m

关 键 词：铝电解 算法 电解温度 深度学习 过程控制 

分 类 号：TH181[机械工程—机械制造及自动化]