Analysis And Mining Of Monitoring Data For Aluminium Reduction Cell

With the continuous development of data measurement technology and means,electrolytic aluminium technicians can have more and more electrolytic cell data which can not be measured online before.Combining new on-line measurement data with daily production data,using artificial intelligence technology,more information of electrolyzer itself can be analyzed and excavated to provide decision-making basis for optimizing production management of electrolyzer,which has important research significance and application value.Firstly,this paper introduces the related data analysis and data mining technology,as well as the performance indicators of various evaluation clustering algorithms.Aiming at the problem that the electrolysis temperature can not be measured on-line,a learning training sample is made up of 20 side-wall temperature data measured by manual daily and minute.Using a large number of side-wall temperature data measured in real time,the electrolysis temperature per hour is predicted by linear regression.Different dimension reduction methods,different linear regression algorithms and cross-validation are discussed.A regression method for predicting electrolysis temperature based on side wall temperature was proposed.In order to solve the problem that traditional K-means algorithm needs to specify K value manually,some performance indexes such as contour coefficient,Hopkins statistics and maximum diameter in clusters are used to dynamically adjust the size of K value in the process of K-means clustering in order to converge to the optimal K value of the system.The design and comparison of K value condensation,splitting and other algorithms are given.Finally,the algorithm is applied to the prediction of electrolysis temperature,the clustering of sidewall temperature and the clustering of anode voltage based on time and space.