Distributed Cell Condition Monitoring Method In Aluminum Electrolysis Based On Dictionary Pair Learning Of Anode Current

The monitoring technology of aluminum electrolytic cell condition plays an important role in the stable operation of aluminum electrolytic cell.Aluminum electrolysis industrial processes are usually highly complex,and the traditional global cell condition monitoring methods ignore the local behaviors of electrolytic cells,resulting in the poor interpretation of monitoring results.The in-depth analysis of the mechanism of aluminum electrolysis process shows that the dynamic behavior of local anode makes the cell condition distributed,and the anode current is closely related to the cell condition.Therefore,a distributed cell condition monitoring method in aluminum electrolysis based on dictionary pair learning of anode current is proposed in this thesis,and the main research contents are summarized as follows:Aiming at the problem of anode current data with high dimension,few labels,dynamic characteristic,noise,and outliers,a dynamic cell condition classification method using unsupervised feature selection based on robust self-representative dictionary pair learning(RSRDPL-UFS)is proposed.The RSRDPL method is used to conduct unsupervised feature selection,and then dynamic classification is performed by the multiple k-nearest neighbor algorithm,providing class labels for subsequent abnormal cell condition monitoring.Aiming at the problems of multi-mode and variables with redundancy and correlation in aluminum electrolysis processes,a distributed abnormal cell condition monitoring method based on joint mutual information and projective dictionary pair learning(JMI-PDPL)is proposed.Firstly,an automatic block division method based on JMI is proposed to divide process variables into several low dimensional blocks.Then,PDPL based monitoring models are built,learning and optimizing dictionaries to calculate reconstruction errors for blocks.Finally,the block statistics are fused to global statistics for mode recognition and anomaly detection,and anomaly sources is further determined with contribution coefficients to achieve anomaly isolation.The experimental results show that: the proposed RSRDPL-UFS method can effectively select the representative features of data to reduce the dimensionality of them.Then,the classification of cell condition and the recognition of new cell condition for reduced-dimension data can be realized by multiple k-nearest neighbor algorithm.The proposed distributed process monitoring method based on JMI-PDPL can provide distributed cell condition monitoring,improve the accuracy of abnormal cell condition monitoring,effectively identify the mode,and availably locate the anomaly source.In addition,to integrate the research achievements of this thesis with the real-world aluminum electrolysis process,a distributed cell condition monitoring system in aluminum electrolysis based on anode current is devised and developed.Figures(38),Tables(16),References(103)...