基于多通道时空卷积注意力网络的铝电解槽阳极效应趋势预测

陈祖国; 刘辉; 黄毅; 陈超洋

doi:10.16383/j.aas.c250635

基于多通道时空卷积注意力网络的铝电解槽阳极效应趋势预测

doi: 10.16383/j.aas.c250635 cstr: 32138.14.j.aas.c260635

陈祖国^{1, 2,},
刘辉^{1, 3,},
黄毅^{1, 2,},
陈超洋^{1, 2,}

1.
湖南科技大学三亚研究院海洋无人探测与控制中心三亚 572000
2.
湖南科技大学信息与电气工程学院湘潭 411201
3.
湖南科技大学计算机科学与工程学院湘潭 411201

基金项目: 国家自然科学基金(62373144, 62403193), 湖南省科技创新计划(2025RC3197, 2024RC9015, 2025RC4012)资助

详细信息

作者简介:
陈祖国：湖南科技大学信息与电气工程学院副教授. 2018年获得中南大学博士学位. 主要研究方向为人工智能, 机器人控制技术和复杂工业过程智能决策. E-mail: zg.chen@hnust.edu.cn

刘辉：湖南科技大学计算机科学与工程学院硕士研究生. 2023年获得广州城市理工学院学士学位. 主要研究方向为工业数据处理. 本文通讯作者. E-mail: 24020502022@mail.hnust.edu.cn

黄毅：湖南科技大学信息与电气工程学院副教授. 2021年获得中南大学博士学位. 主要研究方向为复杂无人系统建模与协同控制, 海洋装备智能化技术, 有色金属冶炼过程优化策略研究. E-mail: yi.huang@hnust.edu.cn

陈超洋：湖南科技大学信息与电气工程学院教授. 2021年获得华中科技大学博士学位. 主要研究方向为智能电网可靠性分析, 复杂网络理论及应用, 群机器人系统协同控制, 网络化系统性能分析, 非线性系统分析与控制. E-mail: ouzk@163.com

计量
- 文章访问数: 18
- HTML全文浏览量: 13
- 被引次数: 0
出版历程
- 收稿日期: 2025-11-17
- 录用日期: 2026-03-19
- 网络出版日期: 2026-04-26

Anode Effect Trend Prediction of Aluminum Electrolysis Cells Based on Multi-channel Spatiotemporal Convolutional Attention Network

CHEN Zhu-Guo^{1, 2
,},
LIU Hui^{1, 3
,},
HUANG Yi^{1, 2
,},
CHEN Chao-Yang^{1, 2
,}

1.
Ocean Unmanned Exploration and Control Center, Sanya Research Institute, Hunan University of Science and Technology, Sanya 572000
2.
College of Information and Electrical Engineering, Hunan University of Science and Technology, Xiangtan 411201
3.
College of Computer Science and Engineering, Hunan University of Science and Technology, Xiangtan 411201

Funds: Supported by National Natural Science Foundation of China (62373144, 62403193) and Technology Innovation Program of Hunan Province (2025RC3197, 2024RC9015, 2025RC4012)

More Information

Author Bio:
CHEN Zu-Guo　 Associate professor at Hunan University of Science and Technology. He received his Ph.D. degree from Central South University in 2018. His research interests include artificial intelligence, robot control technology, and intelligent decision-making for complex industrial processes

LIU Hui　 Master student at the School of Computer Science and Engineering, Hunan University of Science and Technology. He received his bachelor degree from Guangzhou University of Technology in 2023. His main research interest is industrial data processing. Corresponding author of this paper

HUANG Yi　 Associate professor at the School of Information and Electrical Engineering, Hunan University of Science and Technology. He received his Ph.D. degree from Central South University in 2021. His research interests include modeling and cooperative control of complex unmanned systems, intelligent marine equipment technology, and research on optimization strategies for non-ferrous metal smelting processes

CHEN Chao-Yang　 Professor at the School of Information and Electrical Engineering, Hunan University of Science and Technology. He received his Ph.D. degree from Huazhong University of Science and Technology in 2021. His research interests include reliability analysis of smart grids, theory and applications of complex networks, cooperative control of multi-robot systems, performance analysis of networked systems, and analysis and control of nonlinear systems

摘要

摘要: 在铝电解过程中, 阳极效应是影响电能利用效率的典型异常工况, 其发生往往伴随阳极电阻的剧烈波动. 若能实现阳极电阻变化的实时预测, 即可对阳极效应进行前瞻性识别. 为此, 提出一种融合机理约束与数据驱动思想的多通道时空预测模型(卷积长短期记忆–二维卷积–多头注意力, ConvLSTM-Conv2D-MHA), 以联合刻画多槽系统的共性与差异特征. 模型利用堆叠ConvLSTM层提取时序动态, 通过Conv2D分支强化空间特征表达, 并引入MHA机制捕捉长时依赖关系, 从而提升对趋势变化及早期波动的敏感度. 实验结果表明, 该模型在阳极电阻趋势预测中表现出更高的精度与稳定性, 较传统时序模型更能利用多槽间潜在的耦合关联.
- 电阻趋势预测 /
- 多通道ConvLSTM /
- 多头自注意力机制 /
- 阳极效应
Abstract: In the process of aluminum electrolysis, the anode effect is a typical abnormal operating condition that affects energy utilization efficiency and is usually accompanied by sharp fluctuations in anode resistance. Real-time prediction of anode resistance variations enables proactive identification of anode effects. To address this issue, a multi-channel spatiotemporal prediction model integrating convolutional long short-term memory (ConvLSTM), two-dimensional convolution (Conv2D), and multi-head attention (MHA), namely ConvLSTM–Conv2D–MHA, is proposed by combining mechanism constraints with data-driven principles to jointly capture both the commonalities and differences within the multiple system. The model employs stacked ConvLSTM layers to extract temporal dynamics, utilizes a Conv2D branch to enhance spatial feature representation, and incorporates MHA mechanism to capture long-term dependencies, thereby improving sensitivity to trend changes and early-stage fluctuations. Experimental results demonstrate that the proposed model achieves higher accuracy and stability in anode resistance trend prediction and is more effective than conventional time-series models in exploiting the coupling relationships among multiple cells.
- resistance trend prediction /
- multi-channel ConvLSTM /
- multi-head self-attention mechanism /
- anode effect

HTML全文

图 1 ConvLSTM内部结构

Fig. 1 Internal structure of ConvLSTM

下载: 全尺寸图片幻灯片

图 2 多头自注意力机制(MHA)结构

Fig. 2 Structure of multi-head self-attention mechanism (MHA)

下载: 全尺寸图片幻灯片

图 3 CLCM结构

Fig. 3 Structure of CLCM

下载: 全尺寸图片幻灯片

图 4 Conv2D原理流程图

Fig. 4 Flowchart of Conv2D principle

下载: 全尺寸图片幻灯片

图 5 特征融合输出层结构示意图

Fig. 5 Feature fusion output layer structure diagram

下载: 全尺寸图片幻灯片

图 6 基于阳极电阻的阳极效应趋势预测方法流程

Fig. 6 Flowchart of anode effect trend prediction method based on anode resistance

下载: 全尺寸图片幻灯片

图 7 铝电解数据采集与处理系统示意图

Fig. 7 Schematic diagram of aluminum electrolysis data collection and processing system

下载: 全尺寸图片幻灯片

图 8 三槽相关性热力图

Fig. 8 Heatmap of correlation among three cells

下载: 全尺寸图片幻灯片

图 9 CLCM模型预测结果

Fig. 9 Prediction results of CLCM model

下载: 全尺寸图片幻灯片

图 10 单一模型与融合模型的预测结果

Fig. 10 Prediction results of single model and fused model

下载: 全尺寸图片幻灯片

图 11 不同模型局部RMSE随时间变化曲线

Fig. 11 Local RMSE curves of different models over time

下载: 全尺寸图片幻灯片

图 12 不同方法预测性能对比

Fig. 12 Comparison of prediction performance of different methods

下载: 全尺寸图片幻灯片

图 13 不同模型在局部区间的预测结果对比

Fig. 13 Comparison of prediction results of different models in local intervals

下载: 全尺寸图片幻灯片

表 1 模型输入参数

Table 1 Model input parameters

序号	参数	描述
1	槽电压	电解槽电解质熔体中的总电压降
2	滤波电阻	电解槽中过滤系统的电阻
3	平滑电阻	电解槽电解质层的电阻
4	波动差值	电解槽电压或电流波动幅度
5	斜率数据总和	电压或电流变化的速率
6	槽龄	电解槽已运行的时间
7	额外喂料量	电解槽中添加氟盐和三氧化二铝量

下载: 导出CSV

表 2 实验采集的数据信息

Table 2 Information of experimental data collection

参数	数值
数据产生时间	2016年6月25日至 2016年6月28日
电解槽数量/台	3
特征数量/种	7
实验数据样本数/个	34 560
训练集/验证集/测试集占比	28 404 : 3 156 : 3 000

下载: 导出CSV

表 3 CLCM模型参数

Table 3 Parameters of CLCM model

参数	数值
卷积层	3
卷积核大小	1 × 3
移动步长	1
ConvLSTM层隐藏单元	30
初始学习率	0.001
最大迭代次数	100
批量大小batch	64

下载: 导出CSV

表 4 CLCM模型性能评价结果

Table 4 Performance evaluation results of CLCM model

参数	数值
RMSE	8.485
MAPE	0.097%
R²	0.982

下载: 导出CSV

表 5 单一模型和融合模型性能对比

Table 5 Performance comparison between single models and fusion model

模型	RMSE	R²	MAPE
ConvLSTM	9.309	0.978	0.127%
NoMHA	17.995	0.920	0.328%
NoSpatial	15.405	0.940	0.288%
CLCM	8.486	0.982	0.097%

下载: 导出CSV

表 6 消融实验模型局部RMSE显著性检验结果

Table 6 Significance test results of local RMSE for ablation experiment models

对比模型	t值	p值	显著性(p < 0.05)
CLCM与NoMHA	−27.150	< 0.001	显著
CLCM与NoSpatial	−24.163	< 0.001	显著
CLCM与SingleSlot	−22.544	< 0.001	显著

下载: 导出CSV

表 7 不同方法预测评价结果

Table 7 Prediction evaluation results of different methods

模型	RMSE	R²	MAPE
Persistence	22.574	0.873	0.250%
ARIMA	21.273	0.887	0.236%
LSTM_diff	10.537	0.972	0.160%
XGBoost	21.605	0.884	0.405%
TCN	14.422	0.948	0.267%
CLCM	8.486	0.982	0.097%

下载: 导出CSV

表 8 各模型相对于CLCM的DM显著性检验结果

Table 8 DM significance test results of models relative to to CLCM

对比模型	t值	p值	显著性(p < 0.05)
CLCM与LSTM_diff	−11.928	< 0.001	显著
CLCM与ARIMA	−7.482	< 0.001	显著
CLCM与Persistence	−7.919	< 0.001	显著
CLCM与XGBoost	−56.812	< 0.001	显著
CLCM与TCN	−59.111	< 0.001	显著

下载: 导出CSV

参考文献(30)

[1]	桂卫华, 岳伟超, 谢永芳, 张红亮, 阳春华. 铝电解生产智能优化制造研究综述. 自动化学报, 2018, 44(11): 1957−1970 doi: 10.16383/j.aas.2018.c180198 Gui Wei-Hua, Yue Wei-Chao, Xie Yong-Fang, Zhang Hong-Liang, Yang Chun-Hua. Review on intelligent optimization and manufacturing of aluminum electrolysis production. Acta Automatica Sinica, 2018, 44(11): 1957−1970 doi: 10.16383/j.aas.2018.c180198
[2]	Wang J, Xie Y F, Xie S W, Chen X F. Development of data-knowledge-driven predictive model and multi-objective optimization for intelligent optimal control of aluminum electrolysis process. Engineering Applications of Artificial Intelligence, 2024, 134: Article No. 108664 doi: 10.1016/j.engappai.2024.108664
[3]	Yin G, Li Y H, Yan F Y, Quan P C, Wang M, Cao W Q, et al. Deep learning for electrolysis process anode effect prediction based on long short-term memory network and stacked denoising autoencoder. Rare Metals, 2024, 43(12): 6730−6741 doi: 10.1007/s12598-024-02766-x
[4]	叶涛, 朱学峰, 李向阳, 史步海. 基于改进k-最近邻回归算法的软测量建模. 自动化学报, 2007, 33(9): 996−999 doi: 10.16383/j.aas.2007.09.018 Ye Tao, Zhu Xue-Feng, Li Xiang-Yang, Shi Bu-Hai. Soft sensor modeling based on improved k-nearest neighbor regression algorithm. Acta Automatica Sinica, 2007, 33(9): 996−999 doi: 10.16383/j.aas.2007.09.018
[5]	Melo A, Camara M M, Pinto J C. Data-driven process monitoring and fault diagnosis: A comprehensive survey. Processes, 2024, 12(2): Article No. 251 doi: 10.3390/pr12020251
[6]	Li S J, Zhou B H, Shang J L, Chen X F, Yu J B. Review of recent applications and future perspectives on process monitoring approaches in industrial processes. Journal of Manufacturing Systems, 2025, 82: 509−530 doi: 10.1016/j.jmsy.2025.07.002
[7]	Zhang S, Zhang T, Yin Y X, Xiao W D. Alumina concentration detection based on the kernel extreme learning machine. Sensors, 2017, 17(9): Article No. 2002 doi: 10.3390/s17092002
[8]	Xu C H, Tu Z C, Zhang W J, Cen J, Xiong J B, Wang N. A method of optimizing cell voltage based on STA-LSSVM model. Mathematics, 2022, 10(24): Article No. 4710 doi: 10.3390/math10244710
[9]	汤建, 郭海涛, 夏恒, 王鼎, 乔俊飞. 面向工业过程的图像生成及其应用研究综述. 自动化学报, 2024, 50(2): 211−240 doi: 10.16383/j.aas.c230126 Tang Jian, Guo Hai-Tao, Xia Heng, Wang Ding, Qiao Jun-Fei. Image generation and its application research for industrial process: A survey. Acta Automatica Sinica, 2024, 50(2): 211−240 doi: 10.16383/j.aas.c230126
[10]	Li Q S, Zhang Y, Long J H, Bai H, Hu R, Li W. Photovoltaic power prediction based on IMFO-LSTM model. Journal of Electric Power Science and Technology, 2024, 39(3): 199−206
[11]	Xu J Y, Gong J Y, Li Y B, Fu Z, Wang L Y. Surf-riding and broaching prediction of ship sailing in regular waves by LSTM based on the data of ship motion and encounter wave. Ocean Engineering, 2024, 297: Article No. 117010 doi: 10.1016/j.oceaneng.2024.117010
[12]	Lughofer E, Pichler K. Data-driven prediction of possible quality deterioration in injection molding processes. Applied Soft Computing, 2024, 150: Article No. 111029 doi: 10.1016/j.asoc.2023.111029
[13]	Zhao L, Zhang X N, Peng X Y. Power fluctuation mitigation strategy for microgrids based on an LSTM-based power forecasting method. Applied Soft Computing, 2022, 127: Article No. 109370 doi: 10.1016/j.asoc.2022.109370
[14]	He Y L, Lv S H, Zhu Q X, Lu S. Novel multiattribute space-based LSTM for industrial soft sensor applications. IEEE Transactions on Industrial Informatics, 2024, 20(3): 4745−4752 doi: 10.1109/TII.2023.3316289
[15]	Tian H X, Chen J H. Deep learning with spatial attention-based ConvLSTM for SOC estimation of lithium-ion batteries. Processes, 2022, 10(11): Article No. 2185 doi: 10.3390/pr10112185
[16]	张舒, 徐鉴. 时滞耦合系统非线性动力学的研究进展. 力学学报, 2017, 49(3): 565−587 doi: 10.6052/0459-1879-17-123 Zhang Shu, Xu Jian. Review on nonlinear dynamics in systems with coupling delays. Chinese Journal of Theoretical and Applied Mechanics, 2017, 49(3): 565−587 doi: 10.6052/0459-1879-17-123
[17]	Zhang Y D, Jie J, Liu L Q, Lan C M, Shi P. STP-cube-multivariate time series forecasting based on spatiotemporal correlations and periodicity. Neural Processing Letters, 2025, 57(5): Article No. 80 doi: 10.1007/s11063-025-11799-x
[18]	Chen Z G, Lu M, Zhou Y M, Chen C Y. Information-synergy-entropy based multi-feature information fusion for the operating condition identification in aluminium electrolysis. Information Sciences, 2021, 548: 275−294 doi: 10.1016/j.ins.2020.07.031
[19]	Ilyushin Y V, Boronko E A. Analysis of energy sustainability and problems of technological process of primary aluminum production. Energies, 2025, 18(9): Article No. 2194 doi: 10.3390/en18092194
[20]	Sun Y, Gui W H, Chen X F, Xie Y F. Evaluation model of aluminum electrolysis cell condition based on multi-source heterogeneous data fusion. International Journal of Machine Learning and Cybernetics, 2024, 15(4): 1375−1396 doi: 10.1007/s13042-023-01973-9
[21]	da Silva Moreira L J, Besancon G, Ferrante F, Fiacchini M, Roustan H. Convection-diffusion model for alumina concentration in Hall-Héroult process. IFAC-PapersOnLine, 2022, 55(21): 150−155 doi: 10.1016/j.ifacol.2022.09.259
[22]	Wan R Z, Mei S P, Wang J, Liu M, Yang F. Multivariate temporal convolutional network: A deep neural networks approach for multivariate time series forecasting. Electronics, 2019, 8(8): Article No. 876 doi: 10.3390/electronics8080876
[23]	Yuan Q, Wei Z Q, Guan X, Jiang M J, Wang S, Zhang S G, et al. Toxicity prediction method based on multi-channel convolutional neural network. Molecules, 2019, 24(18): Article No. 3383 doi: 10.3390/molecules24183383
[24]	Rosato A, Araneo R, Andreotti A, Succetti F, Panella M. 2-D convolutional deep neural network for the multivariate prediction of photovoltaic time series. Energies, 2021, 14(9): Article No. 2392 doi: 10.3390/en14092392
[25]	Qiu J, Zhang Z D, Zhu Z H, Luo C Q. Accurate state of charge estimation for lithium-ion batteries using a temporal convolutional network and bidirectional long short-term memory hybrid model. Batteries, 2026, 12(2): Article No. 50 doi: 10.3390/batteries12020050
[26]	Ozcan A, Catal C, Kasif A. Energy load forecasting using a dual-stage attention-based recurrent neural network. Sensors, 2021, 21(21): Article No. 7115 doi: 10.3390/s21217115
[27]	Sakib S, Mahadi M K, Abir S R, Moon A M, Shafiullah A, Ali S, et al. Attention-based models for multivariate time series forecasting: multi-step solar irradiation prediction. Heliyon, 2024, 10(6): Article No. e27795 doi: 10.1016/j.heliyon.2024.e27795
[28]	Li S Y, Jin X Y, Xuan Y, Zhou X Y, Chen W H, Wang Y X, et al. Enhancing the locality and breaking the memory bottleneck of Transformer on time series forecasting. In: Proceedings of the 29th Annual Conference on Neural Information Processing Systems. Vancouver, Canada: Advances in Neural Information Processing Systems, 2019. 5244–5254
[29]	Shi X J, Chen Z R, Wang H, Yeung D Y, Wong W K, Woo W C. Convolutional LSTM network: A machine learning approach for precipitation nowcasting. In: Proceedings of the 29th Annual Conference on Neural Information Processing Systems. Montreal, Canada: Advances in Neural Information Processing Systems, 2015. 802–810
[30]	Cui G, Hu T, Zhang W, Bao H J. MMTransformer: A multivariate time-series resource forecasting model for multi-component applications. Scientific Reports, 2025, 15: Article No. 29740 doi: 10.1038/s41598-025-07162-8