Abstract: Data monitoring and control is an important method to improve the quality of production in aluminum electrolysis process.Due to lack of diversity,real-time and stability in data monitoring algorithms,a real-time clustering method for aluminum electrolysis process was studied,and a monitoring model for production singularity based on adaptive Growing Hierarchical Neural Gas (GHNG) was established,in which the mechanisms such as adaptive learning,node generation and deletion and topology display were included.For achieving significant performance boost and improved stability,the cumulative errors of nodes in the model were used to adjust the weights of winning node and its neighborhood nodes adaptively.According to the online data evolution trends,the neural nodes were dynamically deleted and added,and clustering centers were also updated to realize the real-time display of dynamic clustering results,and further the performance of the model and algorithm was tested.An example of aluminum electrolysis process monitoring showed that the proposed algorithm had stronger singularity monitoring capability,especially industrial monitoring and diagnostics applications,and it might assist decision making giving an idea in online quality-control systems for production/managers.

Key words: aluminiumelectrolysis, hierarchical clustering, online monitoring, process control

摘要： 数据监测与控制是铝电解过程提高生产质量的重要手段,针对铝电解过程的数据监测算法缺乏多样性、实时性和稳定性等问题,研究了工艺过程数据实时聚类方法,建立了一种基于自适应生长层次神经气(GHNG)的生产奇异性监测模型。该模型包括自适应学习、节点生成与删除、拓扑结构展示等机制,为提高模型稳定性和分析数据多样性的能力,综合利用节点累积误差自适应调节获胜节点及邻域节点权值;依据在线数据演化趋势动态删除、增加神经节点并更新聚类中心位置,实现实时展现数据实例动态聚类结果,进一步提高聚类算法的时效性,同时对在线监测模型和算法进行了性能测试。最后,通过铝电解过程数据监测实例验证了该模型和算法的奇异性监测能力更强,能对铝电解工艺过程进行准确、有效的监测和控制,为生产/管理者提供决策支持。

关键词: 铝电解, 层次聚类, 在线监测, 过程控制