Research And Application Of Clustering Algorithm For Aluminum Electrolytic Cell Based On DTC

During the aluminum electrolysis production process,a huge amount of data is generated over time.In managing electrolytic cells,process technicians often need to classify cells according to their operating status.If these collected massive data features are extracted and utilized,they can provide decision reference for the daily management of production by electrolysis process technicians to ensure smooth production operation and enhance the intelligent management of aluminum electrolysis production.In this paper,the collected time series production data of aluminum electrolytic cells are studied in depth and a multidimensional time series clustering algorithm is used to cluster and analyze the cells.(1)The raw data of the aluminum electrolyzer over a period of time are collected,the data containing missing values and abnormal values are cleaned and normalized,and then the data are dimensioned down using PCA principal component analysis.Divide the dimensionality-reduced data into fixed-length sub-time series using the sliding window method;(2)For the disadvantages that the grasshopper algorithm has a slow convergence speed and is easy to obtain local optimal solutions,a grasshopper optimization algorithm that incorporates curve adaption and reverse learning is proposed.By using refractive reverse learning to improve population diversity and using a curvilinear adaptive formulation instead of the linear adaptive formulation of the grasshopper algorithm,the convergence speed of the grasshopper algorithm and the ability to obtain globally optimal solutions are improved.The proposed improved method obtained better results by using different test functions for testing.The improved grasshopper algorithm is then used to optimize the k-means algorithm and tested using publicly available datasets and higher accuracy is obtained;(3)A DTC-based clustering model for aluminum electrolytic cells was developed using GRU to replace the LSTM structure in the Deep Temporal Clustering(DTC)model TAE process and an improved k-means algorithm to replace the clustering algorithm in the temporal clustering layer of the DTC model.Experiments show that the DTC-based clustering model for aluminum electrolytic cells works better;(4)The DTC-based aluminum electrolyzer clustering system was designed and developed.The system consists of four modules: data visualization and analysis,data pre-processing,data model training,and data model application.